processes

Review
Application of Machine Learning in Plastic Waste Detection and
Classification: A Systematic Review
Edgar Ramos 1, * , Arminda Guerra Lopes 1 and Fábio Mendonça 2,3

1 Polytechnic Institute of Castelo Branco, Av. Pedro Alvares Cabral 12, 6000-084 Castelo Branco, Portugal;
[email protected]
2 Faculty of Exact Sciences and Engineering, University of Madeira, 9020-105 Funchal, Portugal;
[email protected]
3 Interactive Technologies Institute (ITI/LARSyS and ARDITI), Edif. Madeira Tecnopolo, Caminho da Penteada
Piso-2, 9020-105 Funchal, Portugal
* Correspondence: [email protected]

Abstract: The intersection of artificial intelligence and environmental sustainability has become
a relevant exploration domain in the contemporary era of rapid technological advancements and
complex global challenges. This work reviews the application of machine learning (ML) models
to address the pressing issue of plastic waste (PW) management. By systematically examining the
state of the art with snowballing, this research aims to determine the efficiency and effectiveness of
ML-based methods for PW detection and classification. Considering the increasing environmental
concerns and information processing potential, this article hypothesised that ML models could
contribute to more sustainable PW management practices. For this purpose, two scientific article
repositories were examined from 2000 to 2023, and 188 articles were identified. After the systematic
screening procedure, 28 were selected. Additionally, 28 more articles were included by snowballing.
It was observed that accuracy in either detection or classification problems often exceeded the 80%
detection accuracy benchmark, further improving when the model combination was employed. As a
result, strong support was reached for the applicable potential of ML in PW. It was also concluded
that models based on convolutional neural networks were the most commonly used.
Citation: Ramos, E.; Lopes, A.G.;
Mendonça, F. Application of Machine
Learning in Plastic Waste Detection
Keywords: artificial intelligence; machine learning; plastic waste; recycling; systematic review
and Classification: A Systematic
Review. Processes 2024, 12, 1632.
https://doi.org/10.3390/pr12081632
1. Introduction
Academic Editors: Nediljka
Vukojevic Medvidovic,
The urgent need for environmental sustainability has brought about remarkable global
Ladislav Vrsalović and Emeka efforts to promote recycling on a larger scale. Governments, organisations, and individuals
Emmanuel Oguzie worldwide have acknowledged the importance of recycling to reduce plastic waste (PW)
and conserve resources.
Received: 19 June 2024 Advancements in technology, infrastructure, and policy frameworks have accompa-
Revised: 25 July 2024
nied the evolution of recycling practices. Today, recycling programs encompass a wide
Accepted: 28 July 2024
range of materials, and recycling has become a symbol of hope for our planet. Artifi-
Published: 3 August 2024
cial intelligence (AI) will be vital in addressing challenges and increasing recycling rates
worldwide [1,2].
To explore comprehensively the challenges and opportunities presented by integrating
Copyright: © 2024 by the authors.
machine learning (ML) approaches into PW reduction strategies, we must first characterise
Licensee MDPI, Basel, Switzerland. the context of the problem at hand. The global PW problem is a considerable challenge
This article is an open access article today, casting a shadow over environmental sustainability, economic efficiency, and so-
distributed under the terms and cial well-being. This multi-faceted issue transcends geographical boundaries, affecting
conditions of the Creative Commons communities, industries, and ecosystems worldwide [3].
Attribution (CC BY) license (https:// It was estimated that currently, only 9% of global PW finds its way to effective recycling,
creativecommons.org/licenses/by/ 12% is incinerated, and 79% is accumulated in landfills or in the natural environment. PW,
4.0/). in particular, accounts for a staggering 91% that has yet to be reclaimed [3–5]. These

Processes 2024, 12, 1632. https://doi.org/10.3390/pr12081632 https://www.mdpi.com/journal/processes

Processes 2024, 12, 1632 2 of 19

statistics portend a dire future, with projections indicating that by 2050, oceans may be
burdened with more plastic than fish by weight [6].
The gravity of this environmental crisis necessitates a concerted and innovative response
that uses AI and ML’s transformative potential. These technologies offer promising possibili-
ties for PW reduction and are essential to revolutionising how we approach PW management,
recycling, and resource allocation. It is therefore necessary to explore the complex issue of
global trash and investigate ML-driven solutions to address the problem [7].
The problem of global waste is closely related to the widespread use of plastics in
our daily lives. Although plastics have brought unprecedented convenience, durability,
and versatility to various industries and applications, they have also led to significant
environmental problems.
One of the biggest challenges in waste management (WM) is the increasing use
of non-recyclable plastics. These materials are commonly used for single-use items or
packaging and are difficult to recycle using traditional methods. Their persistence in the
environment worsens pollution and makes it harder to reduce waste in landfills or through
incineration [8].
Efficient WM depends on a strong infrastructure that covers the processes of collection,
sorting, recycling, and disposal. However, many areas worldwide lack the necessary
infrastructure to manage the increasing amount of PW. This inadequacy leads to the
leakage of plastics into natural ecosystems and oceans, causing long-term environmental
damage [9].
Plastics can be classified into seven primary categories based on their formulation.
Each category has a unique Resin Identification Code (RIC) and specific properties affecting
recyclability and disposal methods. For instance, Polyethylene Terephthalate (PETE) is
often used in beverage bottles and food containers and is easy to recycle. On the other hand,
High-Density Polyethylene (HDPE) is known for its strength and is used in containers and
piping. Polyvinyl Chloride (PVC) is commonly used in construction materials, Low-Density
Polyethylene (LDPE) is found in plastic bags, Polypropylene (PP) is used in automotive
parts and textiles, and Polystyrene (PS) is used in insulation and packaging. Finally, the
“Other” category includes various plastics such as polycarbonate and bioplastics. Each
category presents distinct recycling and disposal challenges, requiring nuanced understand-
ing and approaches to manage and mitigate their environmental impact. This diversity
emphasises the complexity of waste management efforts and highlights the importance of
tailored strategies to address each type of plastic’s specific needs and challenges [8].
The problem of how to recycle these was addressed in multiple initiatives worldwide.
Specifically, Operation Green Fences, initiated by Chinese authorities, highlighted the need
for stricter quality controls in the recycling industry. The operation exposed the conse-
quences of lax standards in the global trade of recyclables, illustrating the importance of
responsible recycling practices [10,11]. As expected, solving the PW problem requires inno-
vative technology. AI has emerged with promising tools for reducing waste and increasing
recycling. These technologies can potentially revolutionise PW sorting, classification, and
resource allocation processes, leading to a cleaner and more sustainable planet.
The intersection of AI and environmental sustainability has become a relevant explo-
ration domain in the contemporary era of rapid technological advancements and complex
global challenges. In recent years, substantial progress has been made in applying ML
models to the pressing issue of PW management, with studies showing promising results
in waste detection and classification accuracies. Especially, we can observe a transition
from conventional machine learning models, such as support vector machines, to newer
and more complex models based on deep learning models. These advancements suggest
that ML technologies are approaching the threshold of practical applicability in real-world
waste management contexts, potentially revolutionising how we approach PW manage-
ment, recycling, and resource allocation. Although previous reviews were performed on
general PW, to the best of the authors’ knowledge, a systematic review specific to PW
analysis with detection and classification methodologies was not previously performed.
Processes 2024, 12, 1632 3 of 19

The main objective of this work is to explore the abilities of ML models in developing
efficient strategies for managing PW, which is a growing environmental concern. This article
is based on the concept that ML models can transform the processes of identifying, sorting,
and recycling PW. To examine this proposal, two research questions were formulated:
Can ML models achieve PW detection and classification accuracy suitable for real-world
applications? Which ML approach is more suitable for PW detection?
Guided by this central aim, the article is structured around the following goals: To
identify and examine the leading ML models through a systematic review, highlighting
their capabilities in image-based PW detection and classification. To gather and analyse
existing data to examine the previously identified ML models in PW tasks, establishing a
performance benchmark.
By pursuing these goals, this work seeks to provide a further understanding of the
role of ML in environmental sustainability. This article is structured as follows: Section 2
outlines the search strategy used to identify the initial articles for evaluation. It also covers
the eligibility criteria during the screening stage and concludes with the analysis method
and the articles included. Section 3 describes the search results and thoroughly analyses
the included articles. Section 4 highlights the conclusions of the article by describing the
major findings, while Section 5 outlines the main challenges and future research directions.

2. Materials and Methods

This section presents the methods used for retrieving and analysing articles. We have
performed the analysis based on the Preferred Reporting Items for Systematic Reviews and
Meta-Analyses (PRISMA) guidelines to allow the reproducibility of our results [12].

2.1. Search Strategy

This work searched articles across two databases: Web of Science Core Collection
(WSCC) and Institute of Electrical and Electronics Engineers Xplore (IEEEX). WSCC pro-
vides a comprehensive and multidisciplinary database that offers access to indexed journals
across various fields. The IEEE Xplore Digital Library is a specialised database focusing on
electrical engineering, computer science, and electronics. Cumulatively, these databases
ensured a thorough search due to their extensive coverage of multiple fields and publishers,
allowing for a comprehensive examination of the topic under analysis.
The article search was carried out on 17 December 2023 and was filtered to scan
only the article’s title, abstract, and author-defined keywords published in the 2000–2023
timeframe. The search string “artificial intelligence AND machine learning AND recycling”
was utilised to filter and narrow the search results according to the topic of interest.
The keywords were selected to maximise the retrieval of pertinent information while
minimising the loss caused by adjectival usage. Specifically, the keyword “recycling” was
used to ensure that all search results were related to recycling. In addition, and to ensure
the inclusion of standard ML models, the keywords “machine learning” and “artificial
intelligence” were used together with the “AND” operator.

2.2. Systematic Search

The systematic article selection process is depicted in the PRISMA diagram shown
in Figure 1. A total of 188 articles were found by searching two different databases, and
the specifics are presented in Figure 1. The WSCC had the highest number of publications,
with 120 articles, while 68 articles were found in IEEEX.
A duplicate record elimination process was conducted before passing the articles to
the initial screening phase. Nine articles were removed due to duplication. Then, three
independent scorers evaluated the relevance of each article. The inclusion criteria were
“articles that included ML or general AI applied to WM”. The exclusion criteria were
“articles not focused on WM that did not have either ML or AI mentioned” and “articles
not written in English”.
Processes 2024, 12, 1632 4 of 19

Figure 1. PRISMA flow diagram of the conducted systematic review.

During this procedure, a voting system was employed. Each scorer evaluated the title
and abstract of each article and voted for inclusion, exclusion, or further discussion. Articles
receiving two votes for inclusion were automatically included, while those receiving two
votes for exclusion were automatically excluded. Of the 179 screened articles, 143 were
excluded, thus resulting in 36 articles in the second screening.
Two articles were excluded during the second screening analysis, which involved the
complete article screening process, due to their lack of specificity of waste type unrelated
to PW [13,14]. Another two were excluded because despite having a broad Focus on WM,
they had a limited specificity to plastic WM [15,16]. Finally, four were excluded because
the reports were scattered on diverse topics related to recycling and WM [17–20]. The
selection process resulted in a total of 28 articles that were included in the systematic
review [5,15,21–46].
The resulting 28 studies included in this review all considered AI methods to address
plastic WM, focusing on reduction and recycling. However, there is a considerable difference
among the articles regarding their use of ML models in analysis. Some articles focused on
practical and objective analysis, while others adopted a more holistic approach to evaluate the
current state of the art. To simplify the analysis, we decided to group them based on the ML
models they used and their accuracy in detecting and/or classifying waste.

3. Results and Discussion

Based on an analysis of the publication year of the 28 included articles, it was observed
that research activity on the studied subject began five years ago. This is shown in Figure 2,
Processes 2024, 12, 1632 5 of 19

which displays the distribution of published articles by year. The systematically reviewed
articles started in 2019 (3.57%), and the gradual increase in the number of publications after
that year indicates a growing interest in the topic. Five articles (17.86%) were published in
2020, five articles (17.86%) were published in 2021, ten articles (37.71%) were published in
2022, and seven articles (25.00%) were published in 2023. The peak in 2022 suggests that the
subject has recently attracted significant attention. The prevalence of this trend emphasises
the contemporary significance of the examined subject matter, stressing the need for this
review to consolidate knowledge and point out new directions for future research.

Figure 2. Published articles by year, indicating the number of systematically reviewed articles,
snowballed articles, and the total.

We then checked the cited literature in the articles, snowballing the additional lit-
erature, especially from previous non-systematic reviews [5,23]. These reviews pointed
out the relevance of deep learning-based models for PW detection and classification. The
additional articles were only included if they were published with peer review and if they
explicitly mentioned the usage of the method for plastic recycling. This way, it was possible
to include 28 additional articles [47–74]. These results are presented in Figure 2, showing
that the total number of examined articles was 56.
The following subsections analyse the results summarised in Table 1, which provides
an overview of these articles, including how many classes/categories were considered.
The accuracy is presented in a broader way as an initial examination combining different
methodologies used in different articles. It was observed that most works use a standard
convolutional neural network (CNN) architecture or a model that is based on it, such as
Single-Shot MultiBox Detector (SSD) and You Only Look Once (YOLO). Therefore, two
groups were formed, specifically, the articles that used these ML models and those that
used other approaches. When the used architecture was named, it was opted to specify it;
otherwise, it is just indicated as CNN.
It is important to note that the additional literature included by snowballing was used
only to clarify the performance of the considered models and is not intended to be included
in the in-depth subsequent analysis since the articles were not identified systematically.
Additionally, the articles identified by snowballing (marked with $ and the reference in
front) are included in the same row of the article that was identified by the systematic search
to facilitate the replicability of the search. No duplicated snowballing-identified articles
were included, and the number of classes/categories and year of publication indicated in
the row are unrelated to these articles.
Furthermore, three review articles were included but did not specifically examine the
considered models. Specifically, Seyyedi et al. [37] studied AI-based systems concerning
marine plastics and circular economy. There is also a review by Shennib and Schmitt [31]
regarding circular economy and WM systems using AI and a review on enzyme-embedded
biodegradable agricultural plastics by Maraveas et al. [35]. Several parameters were
extracted from the various articles, and the data from the different articles are presented
in Table 1.
Processes 2024, 12, 1632 6 of 19

A bibliometric analysis was also performed. Of the reviewed 56 articles, 27 were

published in international conferences (13 in the systematically reviewed articles) and 29 in
journals (15 in the systematically reviewed articles). However, conference articles that were
published in a journal paper (for example, the Journal of Physics: Conference Series) were
counted as journal articles. The analysis ended up as a balanced mix between international
conferences and journal publications. Regarding the number of citations indicated in
Google Scholar, for the systematically reviewed articles (on 16 June 2024), it was 1183 for
the systematically reviewed articles and 1693 for the snowballed articles, leading to a total
of 2876 citations, highlighting the relevance of the reviewed topic. Furthermore, for the
systematically reviewed articles, the average number of citations was 36.32 (ranging from
0 to 206), with Sheng et al. [26] work having the highest number of citations (published in
2020), while for the snowballed articles, the average number of citations was 68.93 (ranging
from 5 to 315), with Adedeji and Wang [53] article reaching the highest number of citations
(published in 2019). The average number of citations was 52.63 when we consider both
systematically reviewed articles and snowballed articles. The most commonly referenced
journal was IEEE Access. However, identifying a predominant author was not possible,
indicating that there may not yet be dedicated research lines in this field. Instead, the
publications appear to be more sporadic.
The five most common general keywords in the examined articles (from most to least
common), produced using the pyBibX library (algorithmically generated keywords) on
title and abstract, were deep learning, waste management, machine learning, recycling,
and artificial intelligence. The frequency of these most common keywords is presented
in Figure 3. We can see the top ten general keywords per year of publication in Figure 4,
where the prevalence of “deep learning” after 2019 is apparent, along with the emergence of
keywords “Computers” in 2016, “sociology” in 2020, and “policy-making” in 2023 reflecting
the evolving focus and interdisciplinary nature of research over the years. The graph
presented in Figure 5 was produced by adjacency analysis of the general keywords. It is
highly relevant to note that “deep learning” and “recycling” (the two nodes with more links)
are related to “plastics”. Furthermore, “machine learning” is the third node with more links
and is associated with “recycling” directly or through “neural networks”, while the link to
“deep learning” by “image classification” and “convolutional neural networks”. The ten
most prevalent words (with four or more characters) in the abstract of the examined articles,
presented in Figure 6, were (from most to least common) “waste”, “classification”, “image”,
“model”, “garbage”, “learning”, “plastic”, “management”, “trash”, and “accuracy”. The
most common N-grams with two words were (from most to least common) “deep learning”,
“waste management”, “machine learning”, and “artificial intelligence”. From these results,
it is possible to conclude that the selected articles are indeed aligned with the scope of
the review.

Figure 3. Frequency of the most common general keywords in the examined articles.
Processes 2024, 12, 1632 7 of 19

Figure 4. Frequency of the top ten most common general keywords in the examined articles per
publication year (the three keywords on top of the box in 2016, 2020, and 2023 are used to stress them,
but they refer to the empty box below). The size of the box indicates the frequency of the keyword.

Figure 5. Graph produced by adjacency analysis of the general keywords in the examined articles.

Figure 6. Word cloud produced from the reviewed articles abstracts with the most common words.
Processes 2024, 12, 1632 8 of 19

Table 1. Analysis of the reviewed articles that have reported an accuracy-based metric.

Year of Number of Detection Accuracy Classification Accuracy

Study
Publication Categories/Classes (Specific Method—Accuracy in %) (Specific Method—Accuracy in %)
[25] 2020 4 YOLOv3—85.00 + -
[26] 2020 3 MobileNetV2—86.23 # -
[39] 2022 2 SSD-MobileNet-V1—63.64 # -
[44] 2022 5 EfficientDet—67.40 -
Resnet50—87.00 [53] $
Resnet50—88.00 [52] $
[30] 2023 6 -
VGG19—88.00 [54] $
ConvoWaste—98.30
[24] 2020 - - ResNet-34—89.96
CNN—89.00 [48] $
[45] 2020 2 - CNN—98.20 [47] $
AlexNet and GoogleNet and ResNet-50—99.95
AlexNet—75.00
[22] 2021 6 -
InceptionV1—82.00
Deep CNN4—37.00
[15] 2022 4 -
Deep CNN5—56.70
SqueezeNet—66.84
AlexNet—68.13
InceptionNet—74.41
ResNet—76.59
[43] * 2022 4 - MobileNet_V2—76.77
GoogleNet—76.89
VggNet—77.78
EfficientNet—79.49
DenseNet—80.63
[32] 2022 4 - CNN—95.63
CNN—80.88 [50] $
VGG16—88.42 [49] $
[21] 2022 4 -
MLB-DCNN—92.60 [51] $
FNN-TH—97.02
CNN—95.60
[46] ~ 2022 5 -
YOLOv5—98.30
Resnet-50—96.50
[38] 2022 3 - InceptionV3—98.60
MobileNetV2—99.60
InceptionV3—36.00
ResNeX50—45.20
[42] 2023 4 - VGG-16—46.50
ResNet50—47.85
ResNet—52.44
EfficientNet-B3—97.32
[36] 2023 8 -
CNN—98.50 [55] $
AlexNet—83.00 [74] $
VGG16—93.00 [73] $
CNN—64.00 [71] $
[5] † 2021 - CNN—93.50 [70] $
R-CNN—74.10 [69] $
Capsule-Net—93.60 [72] $
Capsule-Net—95.80 [72] $
YOLOv5—73.20 [59] $
Tiny-YOLO—31.60 [68] $ CNN—92.20 [63] $
YOLOv2—47.90 [68] $ EfficientDet—92.87 [58] $
SSD—67.40 [68] $ InceptionV3—93.13 [60] $
[23] † 2022 - YOLO-Green—78.04 [65] $ AlphaTrash—94.00 [64] $
Faster RCNN—81.00 [68] $ ThanosNet—94.70 [61] $
ResNet-50—81.48 [66] $ GCNet—97.54 [56] $
L-SSD—83.48 [67] $ DNN-TC—98.00 [62] $
DSCAM—98.90 [57] $
* Only the results of the models with pre-training are presented. ~ Only the results of the model with the 80–20%
approach are presented, and we are only reporting the CNN-based models. # Reported as Mean Average Precision
(mAP). + Reported as recognition rate. $ Results of an article identified by snowballing. † Review article.

Figure 7 presents a document projection based on the abstract words. The numbers are
the references of the articles. The projections were performed using the truncated singular
value decomposition method. Embeddings were produced using sentence transformers (all-
MiniLM-L6-v2). From the results of Figure 7, it is noticeable that [29] is the most dissimilar,
Processes 2024, 12, 1632 9 of 19

which is expected since it focuses on circular supply chain hierarchical structures, which
is a topic that is unrelated to the remaining examined articles. Likewise, we can see [35],
which examines enzyme-embedded and microbial degradation methods. [37] is about
marine plastic, and [31] also examines the circular economy concept. Reference [67] is
also separated as it discusses feature fusion concepts, which are still related to the main
research line of other articles but are not prevalent. The big cluster of numbers on the
right is associated with the works that are more similar and are aligned with the scope of
this review.

Figure 7. Document projection of the reviewed articles based on the abstract words. The numbers are
the reference of the articles.

3.1. Studies Developed Without Using the Considered ML Models

Huang and Koroteev [27] proposed an integrated approach that uses AI and ML to
plan and manage energy and waste, including recycling processes. The study involves
the application of neural networks and ML algorithms to predict waste amounts and im-
prove waste collection efficiency. The proposed framework can considerably reduce waste
quantities, landfill use, and transportation needs by applying intelligent WM strategies,
demonstrating a significant potential impact on improving recycling efficiency and WM.
Integrating neural networks for waste prediction and ML algorithms for optimisation in an
energy and WM context represents an innovative and practical approach.
Tseng et al. [29] explore the development of a data-driven circular supply chain (CSC).
Such a structure is crucial for effective resource management and promoting recycling,
both of which are crucial for a cleaner planet. This study highlights the significance of
AI and ML in analysing big data to facilitate better decision-making in CSC. The use
of data-driven tools such as the fuzzy Delphi method, fuzzy decision-making trial and
evaluation laboratory, and entropy weight method demonstrate the application of AI in
enhancing recycling and WM practices. This article provides a comprehensive analysis
using advanced methodologies to understand and optimise the circular supply chain.
The article presented by Shennib and Schmitt [31] thoroughly examines data-driven
technologies and AI applications used in the circular economy and WM systems. The study
investigates different applications of AI in WM, such as product lifecycle management,
waste generation modelling, community engagement, and waste sorting. These areas play
an essential role in advancing recycling technologies and strategies, directly addressing the
subject of our research. The article analyses the current state of data-driven technologies
and AI in WM, identifying gaps and proposing new areas for research and development.
Yu et al. [33] focus on environmental planning with a focus on the principles of
Reduce, Reuse, Recycle, and Recover (4R) supported by an AI-based Hybridised Intelligent
Framework (AIHIF). The goal is to promote smarter solutions for a cleaner planet by
Processes 2024, 12, 1632 10 of 19

applying AI and ML in WM and reducing PW. The article proposes a novel AIHIF for
WM within the 4R concept, aiming to optimise waste collection, promote recycling, and
ensure efficient resource recovery. This approach could considerably enhance recycling
rates and overall WM efficiency, contributing to environmental sustainability and cleaner
urban management.
Carrera et al. [34] present an economic framework for quality sorting control in plastic
recycling classification using ML and spectroscopy technologies. The article incorporates
ML algorithms for classifying plastics based on their infrared spectrum, directly addressing
the AI and ML aspects. The proposed framework utilises Fourier-transform infrared and
near-infrared spectroscopies combined with ML algorithms to classify different types of
plastics. The economic analysis of recycling revenue for various polymers and the selection
of the most economically advantageous algorithms provide an innovative approach to
enhancing the efficiency and profitability of the recycling industry. This could significantly
contribute to developing cost-effective recycling strategies.
The article presented by Maraveas et al. [35] comprehensively reviews enzyme-
embedded and microbial plastics in agricultural use, focusing on environmentally sustain-
able solutions. The focus on enzyme-embedded technologies and microbial degradation
offers a novel perspective on WM strategies.
Seyyedi et al. [37] provide a detailed overview of the challenges and solutions related
to marine plastics in the context of a circular economy, highlighting the crucial role played
by AI in achieving a cleaner planet. The report explores the use of AI-based systems for
managing ocean PW, explores AI models for predicting the accumulation of ocean PW,
and offers insights into policy-making for effective plastic recycling. The article covers
a wide range of topics, including the effects of PW on marine ecology, computational
methodologies utilising AI, and various approaches to manage and reduce marine plastic
pollution.
Kumar and Chimmani [40] examine the application of AI to manage resources and
reduce smart home waste. While it may not specifically mention PW, the broader context
of waste reduction aligns with this article’s analysis (cleaner solutions for the planet).
Finally, Imran et al. [41] present a comprehensive approach to WM using Quantum
Geographic Information Systems (QGIS) for descriptive and predictive data analysis. The
article details predictive analytics, a subset of ML, to forecast waste amounts and optimise
WM operations. It does not explicitly mention AI or recycling, but the principles and
methodologies discussed can be applied to these areas, making it tangentially relevant
to the analysis. The use of QGIS for WM is innovative. It reflects a growing trend of
incorporating geospatial technologies with ML for environmental solutions.

3.2. Studies Based on the Considered ML Models

ML is constantly advancing, and the success of models for practical applications
depends on their accuracy and efficiency. This subchapter provides a detailed analysis
of various ML methods, including their detection accuracy, classification accuracy, and
combined precision. These methods help to improve areas such as object recognition,
detection, semantic segmentation, and instance segmentation.
In waste management, the accuracy of ML methods is essential for waste detection
and classification. Waste detection involves identifying waste within a given environment
with high precision without false positives. ML models like CNNs and YOLOs are trained
to recognise waste objects in complex backgrounds and varying conditions. On the other
hand, waste classification categorises identified waste into specific types or materials like
plastics, organics, or metals. This task requires a deeper analysis of the detected items,
where models like CNNs are further refined to classify the nuanced characteristics of each
waste type.
Classification accuracy is also relevant to effective sorting and recycling processes.
Detection accuracy focuses on correctly identifying waste items, whereas classification
accuracy measures the precision in assigning the correct category to each detected item.
Processes 2024, 12, 1632 11 of 19

Both accuracies address different challenges in the waste management pipeline, with
detection serving as the foundational step and classification as the subsequent detailed
analysis required for effective sorting and recycling.
Kang et al. [24] highlight the significance of an automatic garbage classification system
that employs deep learning techniques. Such systems can be essential for effective WM and
reduction, which contributes to developing smarter solutions for a cleaner planet, especially
for PW recycling. The article explains how deep learning can automatically classify garbage.
The article recommends structural and functional improvements in a deep learning model
to enhance garbage classification. These improvements include multi-feature fusion, feature
reuse, and optimised activation functions. The article reports high classification accuracy
and a quick classification cycle, indicating that this system has the potential to improve
recycling efficiency and WM significantly.
Rahman and Das’s [21] research introduces a novel hybrid deep learning framework
designed to enhance waste classification, an integral component of effective waste manage-
ment and recycling processes. By integrating custom-tailored deep learning architectures,
including CNNs and EfficientNet models, the study significantly contributes to PW reduc-
tion. It leverages advanced AI techniques to accurately categorise various waste types,
particularly plastics, underscoring AI and ML’s pivotal role in driving cleaner, more intelli-
gent solutions for environmental sustainability. The proposed methodology aligns with
the article’s thematic core and promises substantial accuracy enhancements over existing
classification methods.
Alzyoud et al.’s [22] study explores a semi-smart adaptive approach to trash classifica-
tion, blending physical sorting mechanisms with advanced AI techniques, such as CNNs,
to enhance waste management and recycling efficacy. The report presents a comprehensive
strategy that resonates with the pursuit of intelligent waste management solutions by
incorporating methods such as barcode separation, magnetic separators, and hardness tests
alongside CNNs for image classification. The innovative integration of physical and digital
sorting technologies exemplifies the practical application of AI and ML in addressing
environmental challenges and directly contributes to the goal of PW reduction.
Abdu and Noor’s [23] survey paper examines the application of deep learning tech-
nologies in waste detection and classification, offering a broad perspective on the role of AI
and ML in enhancing waste management systems. The article reviews image classification
and object detection models, showcasing their relevance to recycling efforts and efficient
waste management.
Luo et al. [25] introduce an innovative edge-cloud framework that employs Deep
CNNs (DCNNs) and YOLOv3 to integrate edge computing with cloud-based services
for precise image classification and object detection. The proposed system’s capacity to
deliver accurate and rapid detection of recyclable waste while addressing computational
and latency challenges inherent in deep learning applications presents a scalable solution
with significant potential to revolutionise waste management practices.
Sheng et al. [26] explore a waste management system that synergises IoT technol-
ogy, LoRa communication, and a TensorFlow-based deep learning model to detect and
classify waste items, including plastics. Specifically, TensorFlow allows robust deep learn-
ing capabilities in object identification, and LoRa can be used for efficient long-range
communication.
Ozdemir et al.’s [5] paper explores applying ML techniques to recycling, including
CNNs, support vector machines, decision trees, k-nearest neighbours, and standard artifi-
cial neural networks. The paper’s comprehensive analysis provides insights into various
ML applications in recycling, demonstrating how these technologies can considerably
improve waste management outcomes.
Bhattacharya et al.’s [28] study explores deep learning, particularly CNNs, for au-
tomated garbage classification. The introduction of advanced deep learning techniques
to improve sorting accuracy represents an important leap forward in minimising the
inefficiencies associated with manual sorting methods.
Processes 2024, 12, 1632 12 of 19

The paper presented by Nafiz et al. [30] introduces ConvoWaste, a novel automatic
waste segregation system that harnesses CNNs within a deep learning framework to sort
various waste types, including plastics. ConvoWaste exemplifies the practical application of
deep learning in recycling, employing advanced AI to differentiate and accurately classify
waste materials such as plastics, metals, glass, organic substances, medical, and e-waste into
designated categories. Moreover, ConvoWaste innovative use of Capsule-Net for image
classification, combined with a hardware setup involving ultrasonic sensors and servo
motors for the physical sorting of waste, showcases a holistic and advanced approach to
waste segregation. With a reported classification accuracy of 98% and features designed
to notify authorities about waste levels, this system presents a significant leap forward in
waste management technology.
Wu et al. [32] introduce an intelligent dustbin designed to use ML for effective garbage
classification. By employing CNNs for image recognition, the system categorises waste
into distinct groups, such as recyclables, kitchen waste, and harmful materials. Integrating
advanced technologies, including intelligent speech recognition, sensor applications, and
a visual recognition system, into the dustbin design marks a relevant step forward in
waste management technology, potentially leading to higher recycling rates. The system’s
capability to upload classification data to the cloud further opens the possibility for data
analysis and system optimisation.
Altikat et al.’s [15] study examines various DCNN architectures to accurately iden-
tify and categorise different waste types. Despite the challenges posed by the inherent
properties of PW, such as transparency and deformation, the paper’s approach to tun-
ing DCNN models to improve waste classification accuracy highlights the potential of
AI-driven solutions to overcome obstacles in waste segregation.
Tripathi et al. [36] introduce a novel application of the EfficientNet-B3 CNN to waste
material classification. With a notable accuracy rate of 97%, the system shows the potential
of deep learning algorithms to improve waste segregation.
Zia et al. [38] introduce a reverse vending machine (RVM) designed for efficient PW
management, using MobileNet to classify plastic bottles precisely. The RVM’s design,
characterised by its affordability, portability, and high classification accuracy, introduces a
new, user-friendly approach to waste management. The machine’s success in a university
setting, evidenced by substantial PW collection, underscores its effectiveness.
The PLEESE system, introduced by Salim et al. [39], represents a novel intersection of
technology and environmental stewardship. It aims to promote the reuse of plastic items
by using computer vision and deep learning. By identifying reusable plastic containers
through ML algorithms, PLEESE exemplifies the practical application of AI technologies
in the promotion of sustainable behaviours. While the system’s primary focus is on the
reuse aspect, it indirectly encourages recycling efforts by minimising the volume of waste
requiring processing. The innovative nature of PLEESE lies in its strategy to effect be-
havioural change at the point of disposal, utilising persuasive messaging based on deep
learning-driven identification of reusable plastics. This proactive stance on waste manage-
ment, particularly suited for high-visibility areas like urban centres, has the potential to
alter individual behaviours towards plastic use and contributes significantly to the aim of
mitigating plastic pollution.
Y. Pan’s [42] study examines domestic garbage classification through the lens of
deep learning, presenting methodologies that, while not exclusively focused on PW, are
highly pertinent to the broader objectives of enhancing waste management and recycling
processes. The exploration of image classification algorithms, particularly through models
like ResNet, highlights the application of advanced AI and ML techniques in identifying and
categorising waste materials. While the paper’s primary discourse centres around domestic
garbage, the implications for PW management are implicit and significant, suggesting that
the methodologies discussed could be seamlessly adapted to target PW specifically.
Liang and Gumabay’s [43] study introduces a smart household waste classification
system that uses AI to enhance waste sorting and management. While the primary focus
Processes 2024, 12, 1632 13 of 19

is household waste, the AI methodologies employed, including CNNs, hold significant

promise for application in general PW management.
The paper presented by Chopde et al. [44] develops an AI-driven system that au-
tomatically classifies waste (using computer vision techniques), including plastics. The
system operates on a compact and accessible platform, utilising a Raspberry Pi and a cam-
era module, and employs the EfficientDet model for object detection. This methodology
demonstrates the feasibility of integrating AI models into everyday waste management
tools. It highlights the potential for such technologies to considerably improve the precision
and efficiency of waste classification locally. The implications of deploying this technology
are substantial, promising to elevate the effectiveness of recycling programs.
Togacar et al.’s [45] study introduces an approach to waste classification that integrates
autoencoder networks with feature selection techniques in CNN models, achieving a
classification accuracy of 99.95%. This level of performance in distinguishing between
various waste types holds immense potential for improving the sorting and recycling of
materials, including plastics.
Finally, Ghatkamble et al.’s [46] study introduces an intelligent municipal waste
management system that uses the YOLO network alongside IoT technology. This system’s
focus is on employing AI techniques for categorising and efficiently managing municipal
waste. The innovative fusion of YOLO networks with IoT technology for real-time waste
management underscores a novel approach in the domain, offering potential advancements
in the efficiency and intelligence of waste handling systems.

3.3. Performance Analysis

In this subsection, we will conduct an analysis of the accuracy of various ML models
in object detection and classification. Of the systematically reviewed studies, 16 examined
the considered CNN-based models, with their performance being summarised in Table 2.
The additional snowballed literature and the combination of both (indicated as global) are
also included in the table. It is important to notice that the studies used different databases,
training conditions, model structures, and number of classes/categories. Thus, this analysis
can only be seen as an initial approximation, highlighting the tendencies regarding accuracy
in detection and classification. Therefore, it serves as a global overview of the trends. Thus,
the rationale is to provide this overview rather than an in-depth examination, which would
have required running all models on the same standard dataset.

Table 2. Overview of the performance of the considered ML models.

Detection Accuracy Detection Accuracy Classification Classification Accuracy

ML Models
Data Points Average in % Accuracy Data Points Average in %
Total and weighted average 5 80.11 28 75.36
Snowballed total and average 12 71.36 20 92.05
Global total and average 17 74.86 48 82.62

Regarding the number of categories/classes used by the systematically reviewed

articles, it ranged from 2 to 8, with a median of 4 (average of 4.27). More specifically, for
detection problems, the number of categories used ranged from 2 to 6, with a median and
average of 4. On the other hand, for the classification problems, the number of categories
used ranged from 2 to 8, with a median of 4 (average of 4.40). These results suggest the
trend of using four categories/classes.
By examining Tables 1 and 2, it is apparent that the accuracy in detection problems
varies substantially, reflecting the difficulties associated with the need to extract spatial
features from images effectively. SSD models are optimised for real-time detection tasks,
balancing speed and performance adeptly, and can surpass an accuracy of 84%. YOLO
models, which reached a detection accuracy of 85%, are known for their efficiency in
Processes 2024, 12, 1632 14 of 19

processing images in a single evaluation pass, providing rapid and accurate detections. It is
relevant to notice the global average accuracy of around 75% for the 17 examined samples.
Shifting to classification accuracy, the data highlights the adaptability and effectiveness
of CNN in a wide range of image classification tasks, achieving an average global accuracy
of around 83%. This reinforces the position of CNNs as versatile tools in ML.
Although SSD primarily focuses on object detection, it also shows potential for classifica-
tion tasks, with an accuracy of 76.77% [43]. On the other hand, YOLO stands out in classification
with an impressive accuracy of 98.05% [46]. However, further research is necessary to validate
the capabilities of these models as the representation is from a single article.
Custom models designed for specific applications perform exceptionally well, with
an accuracy surpassing 95% [5,23], highlighting the effectiveness of tailored solutions in
achieving high performance. Additionally, combining multiple models has proved success-
ful in classification tasks, achieving an accuracy that surpasses 99% [45], emphasising the
advantage of collaborative approaches in complex ML challenges.
The convergence of the models’ performance towards a substantially high average
detection accuracy is an important milestone in the evolution of detection models, as it
showcases their refined ability to interpret complex and diverse visual data landscapes.
The same is true for classification accuracy, demonstrating the robustness and adaptability
of these models in distinguishing and categorising diverse objects within images. These
accuracy-based metrics signify a broader trend towards performance and reliability, driven
by the contributions from the examined models, from standard CNN architectures to
custom and combined methodologies. YOLO’s notable performance in classification further
enriches this narrative, suggesting an expanding horizon for models traditionally associated
with detection tasks.
Figure 8 shows a boxplot that summarises the central tendencies and variability of
accuracies attained by different ML models. This graphical representation offers a concise
yet comprehensive overview of the data, illustrating the distribution and asymmetry of
accuracy values and pinpointing any outliers.

Figure 8. Boxplot of the performance of the considered ML models.

Considering the classification accuracy of the systematically reviewed article, it is

apparent in Figure 8 that the classification accuracy boxplot displays a wide range of model
performance, with the lowest accuracy recorded at 36% [42]. This suggests that some
models may struggle with complex detection when considering four classes. There are
no minimum outliers, which indicates a consistent performance floor across all models
studied. The middle 50% of data points are delineated by the first quartile (Q1, 25th
percentile) at 59.24% and the third quartile (Q3, 75th percentile) at 96.28%. This signifies
a noteworthy improvement in model performance within this interquartile range, from
moderately performing models to those achieving high accuracy. The median accuracy is
77.34%, slightly exceeding the average accuracy (75.36%); thus, the results are reasonably
well distributed. The maximum accuracy recorded was 99.60% [38], demonstrating the
potential of state-of-the-art models to achieve near-perfect detection in controlled conditions
Processes 2024, 12, 1632 15 of 19

or specific applications. Curiously, the snowballed articles exhibit a much shorter variation,
with an outlier at 73.20% [59]. Q1 is 89.80%, and Q3 is 97.11%, with a median of 93.32%.
Therefore, in the global classification, reference [42]’s performance is now an outlier; Q1 is
75.40%, and Q3 is 96.33%, with a median of 91.08%.
Regarding the detection performance, it is clear from Figure 8 that a lower interquartile
range was attained. For the systematically reviewed articles, Q1 was 65.52%, Q3 was
92.27%, and the median was 85%, which is considerably higher than the mean (80.11%).
This suggests a slight left skew in the data, where most models cluster around a higher
performance band, with fewer models trailing in the lower accuracy regions. This tendency
is even more notorious in the examined snowballed articles, where the median and mean
are 79.52% and 71.36, respectively, with Q1 and Q3 of 64.85% and 86.12, respectively.
The lowest reported performance, comprising an outlier, was 31.60% [68], while the best
performance was 98.30% [30]. Regarding the global detection accuracy, Q1, Q3, and median
were 65.70%, 86.62%, and 81%, respectively.
These results highlight a tendency towards good detection and classification perfor-
mance in the current state of the art. It is especially relevant to the 97.32% classification
accuracy attained by Tripathi et al. [36] using eight classes and the 98.30% detection accu-
racy of ConvoWaste, proposed by Nafiz et al. [30], using six categories. Both articles were
published in 2023, showing increased performance even in the more challenging problems
with a higher number of categories/classes.

4. Conclusions
This systematic review has comprehensively examined the application of ML tech-
niques in PW detection and classification, revealing relevant advancements in the field over
the past five years. Furthermore, a bibliometric analysis was executed to examine relevant
keywords, assess which articles are more similar to each other, and assess the citations of
the reviewed articles, which is a standard examination that was previously conducted on
wastewater treatment with AI [75]. Our analysis demonstrates that CNN-based models,
particularly YOLO and SSD architectures, have emerged as highly effective tools in this
domain, consistently achieving detection accuracies exceeding 80% and classification accu-
racies surpassing 83% across diverse studies. Thus, addressing the research questions, it
was concluded that ML models achieve a detection accuracy that meets but often exceeds
the 80% detection accuracy benchmark. It is also important to notice that in many instances,
the accuracy surpassed the 95% threshold. These results support a positive answer to the
first research question, “Can ML models achieve PW detection and classification accuracy
suitable for real-world applications?”. Regarding the second research question, “Which
ML approach is more suitable for PW detection?”, it is more challenging to provide a more
direct answer; however, the results of Table 1 suggest that YOLO is likely the best model.
The field has witnessed rapid progress, with recent studies in 2023 reporting classifi-
cation accuracies of 97.32% for eight waste categories and detection accuracies of 98.30%
for six categories, indicating the growing capability of these systems to handle complex,
multi-class waste management scenarios. This progress suggests that machine learning
technologies are approaching the threshold of practical applicability in real-world waste
management contexts.
However, significant challenges remain. The variability in methodologies, datasets,
and performance metrics across studies hampers direct comparisons and standardised
benchmarking. Furthermore, while laboratory results are promising, there is a notable gap
between controlled experimental performance and the robustness required for real-world
waste management applications.
These findings have important implications for environmental sustainability efforts
and waste management policies. The high accuracies achieved suggest that ML could
significantly enhance the efficiency and effectiveness of PW sorting and recycling processes,
potentially revolutionising waste management practices. However, successful implementa-
tion will require close collaboration between specialised technicians, waste management
Processes 2024, 12, 1632 16 of 19

professionals, and policymakers to address practical deployment challenges. It is important

to consider the hardware requirements and the need for interpretability [76].

5. Challenges and Future Research Directions

The recommendation of this review for future research is first to develop a benchmark
dataset that allows the examination of all models in the same condition, facilitating direct
comparisons between different approaches. Furthermore, it is essential that the used
datasets are made publicly available to allow independent validation of the models and
results. Lastly, validating the performance and scalability of the proposed solutions in
real-world pilot studies is also needed.
Future research lines could examine multimodal approaches by combining image data
with other sensor types to improve robustness in varied environmental conditions, explore
federated learning techniques to enable collaborative model training, and examine the
potential of unsupervised and self-supervised learning methods to reduce reliance on large
labelled datasets. It is also advisable to further refine the transfer learning and few-shot
learning methods that were used to address the challenge of limited labelled data in specific
waste management contexts.
Subsequent reviews should address several key areas identified in this work. As the
field matures and more empirical studies become available, a more granular subgroup
analysis examining factors such as dataset size, waste category count, and algorithmic
variations could provide insights into performance differentials across studies. A compre-
hensive temporal analysis of model performance evolution would indicate the trajectory
and rate of progress within the domain. Furthermore, examining the computational re-
source requirements associated with various approaches would clarify the feasibility of
real-world implementation and provide scalability considerations. From an economic
perspective, as these methodologies transition from theoretical constructs to practical ap-
plications, rigorous assessments of the economic viability of deploying these technologies
across diverse contexts become essential. These assessments would provide valuable data
to inform decision-making processes for potential adopters and policymakers.

Author Contributions: Conceptualization, E.R., A.G.L. and F.M.; methodology, E.R., A.G.L. and F.M.;
software, E.R.; validation, A.G.L. and F.M.; formal analysis, E.R.; investigation, E.R., A.G.L. and F.M.;
resources, E.R.; data curation, E.R.; writing—original draft preparation, E.R.; writing—review and
editing, A.G.L. and F.M.; visualization, E.R.; supervision, A.G.L. and F.M.; project administration,
A.G.L. and F.M.; funding acquisition, A.G.L. and F.M. All authors have read and agreed to the
published version of the manuscript.
Funding: ITI/Larsys—Funded by FCT projects 10.54499/LA/P/0083/2020, 10.54499/UIDP/50009/2020,
and 10.54499/UIDB/50009/2020.
Data Availability Statement: The original contributions presented in the study are included in the
article, further inquiries can be directed to the corresponding author.
Conflicts of Interest: The authors declare no conflicts of interest.

References
1. Environment Programme. We’re Gobbling up the Earth’s Resources at an Unsustainable Rate. Available online: https://www.
unep.org/news-and-stories/story/were-gobbling-earths-resources-unsustainable-rate (accessed on 17 December 2023).
2. d’Ambrières, W.; Plastics Recycling Worldwide: Current Overview and Desirable Changes. December 2019. Available online:
https://journals.openedition.org/factsreports/5102 (accessed on 17 December 2023).
3. Geyer, R.; Jambeck, J.R.; Production, K.L.L. Production use, and fate of all plastics ever made. Sci. Adv. 2017, 3, e1700782.
[CrossRef] [PubMed]
4. Smulian, M. OECD Finds Only 9% of Plastic Recycled Worldwide. Available online: https://www.mrw.co.uk/news/oecd-finds-
only-9-of-plastic-recycled-worldwide-22-02-2022/ (accessed on 17 December 2023).
5. Ozdemir, M.E.; Ali, Z.; Subeshan, B.; Asmatulu, E. Applying machine learning approach in recycling. J. Mater. Cycles Waste
Manag. 2021, 23, 855–871. [CrossRef]
6. Garello, R. Plastics: A Threat to our Oceans. Available online: https://ieeeoes.org/oes-beacon/june-2019-beacon/plastics-a-
threat-to-our-oceans/ (accessed on 17 December 2023).
Processes 2024, 12, 1632 17 of 19

7. Jason, C. AI-Guided Robots Are Ready to Sort Your Recyclables. Available online: https://spectrum.ieee.org/ai-guided-robots-
are-ready-to-sort-your-recyclables (accessed on 17 December 2023).
8. Chidepatil, A.; Bindra, P.; Kulkarni, D.; Qazi, M.; Kshirsagar, M.; Sankaran, K. From Trash to Cash: How Blockchain and
Multi-Sensor-Driven Artificial Intelligence Can Transform Circular Economy of Plastic Waste? Adm. Sci. 2020, 10, 23. [CrossRef]
9. Grand View Research. Plastic Waste Management Market Size, Share & Trends Analysis Report by Service (Collection, Recycling,
Incineration, Landfills), by Polymer (PP, PET, PVC), by Source, by End Use, by Region, and Segment Forecasts, 2020–2027.
Available online: https://www.grandviewresearch.com/industry-analysis/plastic-waste-management-market (accessed on 17
December 2023).
10. Flower, W. What Operation Green Fence Has Meant for Recycling. Available online: https://www.waste360.com/waste-
management-business/what-operation-green-fence-has-meant-for-recycling (accessed on 17 December 2023).
11. Recycling Today. China Announces Import Ban on an Additional 32 Scrap Materials. 2023. Available online: https://www.
recyclingtoday.com/article/china-bans-solid-waste-imports/ (accessed on 17 December 2023).
12. Page, M.J.; McKenzie, J.E.; Bossuyt, P.M.; Boutron, I.; Hoffmann, T.C.; Mulrow, C.D.; Shamseer, L.; Tetzlaff, J.M.; Akl, E.A.;
Brennan, S.E.; et al. The PRISMA 2020 statement: An updated guideline for reporting systematic reviews. Int. J. Surg. 2021, 88,
105906. [CrossRef] [PubMed]
13. Hua, D.; Gao, J.; Mayo, R.; Smedley, A.; Puranik, P.; Zhan, J. Segregating Hazardous Waste Using Deep Neural Networks in
Real-Time Video. In Proceedings of the 2020 10th Annual Computing and Communication Workshop and Conference (CCWC),
Las Vegas, NV, USA, 6–8 January 2020; Charkrabarti, S., Paul, R., Eds.; IEEE: Piscataway, NJ, USA, 2020; pp. 1016–1022. [CrossRef]
14. Kumar, N.M.; Mohammed, M.A.; Abdulkareem, K.H.; Damasevicius, R.; Mostafa, S.A.; Maashi, M.S.; Chopra, S.S. Artificial
intelligence-based solution for sorting COVID related medical waste streams and supporting data-driven decisions for smart
circular economy practice. Process Saf. Environ. Prot. 2021, 152, 482–494. [CrossRef]
15. Altikat, A.; Gulbe, A.; Altikat, S. Intelligent solid waste classification using deep convolutional neural networks. Int. J. Environ.
Sci. Technol. 2022, 19, 1285–1292. [CrossRef]
16. Menaka, S.; Gayathri, A. To Improving the Performance of Identification and Segregation of Liquid and Solid from Municipal
Waste Using Adam Optimization Algorithm. In Proceedings of the 2023 International Conference on Self Sustainable Artificial
Intelligence Systems (ICSSAS), Erode, India, 18–20 October 2023; pp. 1515–1520. [CrossRef]
17. Thibuy, K.; Thokrairak, S.; Jitngernmadan, P. Holistic Solution Design and Implementation for Smart City Recycle Waste
Management Case Study: Saensuk City. In Proceedings of the 2020—5th International Conference on Information Technology
(InCIT), Chonburi, Thailand, 21–22 October 2020; pp. 233–237. [CrossRef]
18. Viswanadham, Y.K.; Rani, R.S.; Mai, V.M.S.L.S.; Krishna, V.N.V.; Vamsi, Y.V. An Identification and Categorization of Plastic
Material using Deep Learning Approach. In Proceedings of the 2023 3rd International Conference on Pervasive Computing and
Social Networking (ICPCSN), Salem, India, 19–20 June 2023; pp. 289–294. [CrossRef]
19. Tamin, O.; Moung, E.G.; Dargham, J.A.; Yahya, F.; Omatu, S.; Angeline, L. Machine Learning for Plastic Waste Detection:
State-of-the-art, Challenges, and Solutions. In Proceedings of the 2022 International Conference on Communications, Information,
Electronic and Energy Systems (CIEES), Veliko Tarnovo, Bulgaria, 24–26 November 2022; pp. 1–6. [CrossRef]
20. Shravani, N.; Vigneshwari, S. Optimizing Waste Management through Intelligent Machine Learning Systems: A Comparative
Study. In Proceedings of the 2023 14th International Conference on Computing Communication and Networking Technologies
(ICCCNT), Delhi, India, 6–8 July 2023; pp. 1–6. [CrossRef]
21. Rahman, N.; Das, S.K. A Fusion of Three Custom-Tailored Deep Learning Architectures for Waste Classification. In Proceedings
of the 2022 4th International Conference on Sustainable Technologies for Industry 4.0 (STI), Dhaka, Bangladeshc, 17–18 December
2022; pp. 1–6. [CrossRef]
22. Alzyoud, F.Y.; Maqableh, W.; Al Shrouf, F. A Semi Smart Adaptive Approach for Trash Classification. Int. J. Comput. Commun.
Control 2021, 16, 4172. [CrossRef]
23. Abdu, H.; Noor, M.H.M. A Survey on Waste Detection and Classification Using Deep Learning. IEEE Access 2022, 10, 128151–
128165. [CrossRef]
24. Kang, Z.; Yang, J.; Li, G.; Zhang, Z. An Automatic Garbage Classification System Based on Deep Learning. IEEE Access 2020, 8,
140019–140029. [CrossRef]
25. Luo, Q.; Yang, G.; Zhao, X. An Edge-Cloud Framework Equipped with Deep Learning Model for Recyclable Garbage Detection.
In Proceedings of the 2020 Eighth International Conference on Advanced Cloud and Big Data (CBD), Taiyuan, China, 5–6
December 2020; pp. 248–252. [CrossRef]
26. Sheng, T.J.; Islam, M.S.; Misran, N.; Baharuddin, M.H.; Arshad, H.; Islam, R.; Chowdhury, M.E.H.; Rmili, H. An Internet of Things
Based Smart Waste Management System Using LoRa and Tensorflow Deep Learning Model. IEEE Access 2020, 8, 148793–148811.
[CrossRef]
27. Huang, J.; Koroteev, D.D. Artificial intelligence for planning of energy and waste management. Sustain. Energy Technol. Assess.
2021, 47, 101426. [CrossRef]
28. Bhattacharya, S.; Sai, K.B.; Puvirajan, H.S.; Peera, H.; Jyothi, G. Automated Garbage Classification using Deep Learning. In
Proceedings of the 2023 2nd International Conference on Applied Artificial Intelligence and Computing (ICAAIC), Salem, India,
4–6 May 2023; pp. 404–410. [CrossRef]
Processes 2024, 12, 1632 18 of 19

29. Tseng, M.-L.; Ha, H.M.; Tran, T.P.T.; Bui, T.-D.; Chen, C.-C.; Lin, C.-W. Building a data-driven circular supply chain hierarchical
structure: Resource recovery implementation drives circular business strategy. Bus. Strategy Environ. 2022, 31, 2082–2106. [CrossRef]
30. Nafiz, M.S.; Das, S.S.; Morol, M.K.; Al Juabir, A.; Nandi, D. ConvoWaste: An Automatic Waste Segregation Machine Using Deep
Learning. In Proceedings of the 2023 3rd International Conference on Robotics, Electrical and Signal Processing Techniques
(ICREST), Dhaka, Bangladesh, 7–8 January 2023; pp. 181–186. [CrossRef]
31. Shennib, F.; Schmitt, K. Data-driven technologies and artificial intelligence in circular economy and waste management systems:
A review. In Proceedings of the 2021 IEEE International Symposium on Technology and Society (ISTAS), Waterloo, ON, Canada,
28–31 October 2021; pp. 1–5. [CrossRef]
32. Wu, H.; Zhang, H.; Liang, J.; Yang, Z.; Wang, X. Design of New Intelligent Dustbin for Garbage Classification based on Machine
Learning. In Proceedings of the 2022 International Joint Conference on Information and Communication Engineering (JCICE),
Seoul, Republic of Korea, 20–22 May 2022; pp. 171–176. [CrossRef]
33. Yu, K.H.; Zhang, Y.; Li, D.; Montenegro-Marin, C.E.; Kumar, P.M. Environmental planning based on reduce, reuse, recycle and
recover using artificial intelligence. Environ. Impact Assess. Rev. 2021, 86, 106492. [CrossRef]
34. Carrera, B.; Mata, J.B.; Pinol, V.L.; Kim, K. Environmental sustainability: A machine learning approach for cost analysis in plastic
recycling classification. Resour. Conserv. Recycl. 2023, 197, 107095. [CrossRef]
35. Maraveas, C.; Kotzabasaki, M.I.; Bartzanas, T. Intelligent Technologies, Enzyme-Embedded and Microbial Degradation of
Agricultural Plastics. Agriengineering 2023, 5, 85–111. [CrossRef]
36. Tripathi, R.; Shetty, H.; Patil, K.; Ingawale, P.; Trivedi, M. Intelligent Waste Material Classification Using EfficientNet-B3
Convolutional Neural Network for Enhanced Waste Management. In Proceedings of the 2023 International Conference on Self
Sustainable Artificial Intelligence Systems (ICSSAS), Erode, India, 18–20 October 2023; pp. 132–137. [CrossRef]
37. Seyyedi, S.R.; Kowsari, E.; Ramakrishna, S.; Gheibi, M.; Chinnappan, A. Marine plastics, circular economy, and artificial intelligence:
A comprehensive review of challenges, solutions, and policies. J. Environ. Manag. 2023, 345, 118591. [CrossRef] [PubMed]
38. Zia, H.; Jawaid, M.U.; Fatima, H.S.; Hassan, I.U.; Hussain, A.; Shahzad, S.; Khurram, M. Plastic Waste Management through the
Development of a Low Cost and Light Weight Deep Learning Based Reverse Vending Machine. Recycling 2022, 7, 70. [CrossRef]
39. Salim, A.A.; Karnadi, I.; Williamdy, W.; Bastiaan, T.; Steven, E. PLEESE—Plastic Reuse Reminder System using Computer Vision
Deep Learning. In Proceedings of the 2022 1st International Conference on Technology Innovation and Its Applications (ICTIIA),
Tangerang, Indonesia, 23 September 2022; pp. 1–6. [CrossRef]
40. Kumar, J.N.A.; Chimmani, S. Proposal of smart home resource management for waste reduction and sustainability using AI and
ML. In Proceedings of the 2019 International Conference on Communication and Electronics Systems (ICCES), Coimbatore, India,
17–19 July 2019; pp. 992–998. [CrossRef]
41. Imran; Ahmad, S.; Kim, D.H. Quantum GIS Based Descriptive and Predictive Data Analysis for Effective Planning of Waste
Management. IEEE Access 2020, 8, 46193–46205. [CrossRef]
42. Pan, Y. Research on image classification algorithm of domestic garbage based on deep learning method. In Proceedings of the
2023 IEEE 3rd International Conference on Power, Electronics and Computer Applications (ICPECA), Shenyang, China, 29–31
January 2023; pp. 1237–1244. [CrossRef]
43. Liang, Z.; Gumabay, M.V.N. Smart Household Waste Classification System using Artificial Intelligence. In Proceedings of the
2022 5th International Conference on Advanced Electronic Materials, Computers and Software Engineering (AEMCSE), Wuhan,
China, 22–24 April 2022; pp. 766–771. [CrossRef]
44. Chopde, A.; Bharate, N.; Bhattar, S.; Kunvar, A.; Bhadwal, S. Trash Can! An AI system for automatic classification of waste. In
Proceedings of the 2022 International Conference on Computational Intelligence and Sustainable Engineering Solutions (CISES),
Greater Noida, India, 20–21 May 2022; pp. 167–171. [CrossRef]
45. Togacar, M.; Ergen, B.; Comert, Z. Waste classification using AutoEncoder network with integrated feature selection method in
convolutional neural network models. Measurement 2020, 153, 107459. [CrossRef]
46. Ghatkamble, R.; Parameshachari, B.D.; Pareek, P.K. YOLO Network Based Intelligent Municipal Waste Management in Internet
of Things. In Proceedings of the 2022 Fourth International Conference on Emerging Research in Electronics, Computer Science
and Technology (ICERECT), Mandya, India, 26–27 December 2022; pp. 1–10. [CrossRef]
47. Chu, Y.; Huang, C.; Xie, X.; Tan, B.; Kamal, S.; Xiong, X. Multilayer hybrid deep-learning method for waste classification and
recycling. Comput. Intell. Neurosci. 2018, 2018, 5060857. [CrossRef]
48. Wang, Y.; Zhang, X. Autonomous garbage detection for intelligent urban management. MATEC Web Conf. 2018, 232, 01056.
[CrossRef]
49. Faria, R.; Ahmed, F.; Das, A.; Dey, A. Classification of organic and solid waste using deep convolutional neural networks. In
Proceedings of the 2021 IEEE 9th Region 10 Humanitarian Technology Conference (R10-HTC), Bangalore, India, 30 September–2
October 2021; pp. 1–6.
50. Nnamoko, N.; Barrowclough, J.; Procter, J. Solid waste image classification using deep convolutional neural network. Infrastruc-
tures 2022, 7, 4. [CrossRef]
51. Shi, C.; Tan, C.; Wang, T.; Wang, L. A waste classification method based on a multilayer hybrid convolution neural network. Appl.
Sci. 2021, 11, 18. [CrossRef]
52. Narayan, Y. DeepWaste: Applying Deep Learning to Waste Classification for a Sustainable Planet. In Proceedings of the 34th
Conference on Neural Information Processing Systems (NeurIPS 2020), Vancouver, BC, Canada, 6–12 December 2020.
Processes 2024, 12, 1632 19 of 19

53. Adedeji, O.; Wang, Z. Intelligent Waste Classification System Using Deep Learning Convolutional Neural Network. Procedia
Manuf. 2019, 35, 607–612. [CrossRef]
54. Ramsurrun, N.; Suddul, G.; Armoogum, S.; Foogooa, R. Recyclable Waste Classification Using Computer Vision and Deep
Learning. In Proceedings of the 2021 Zooming Innovation in Consumer Technologies Conference (ZINC), Novi Sad, Serbia, 26–27
May 2021; pp. 11–15.
55. Yuan, Z.; Liu, J. A Hybrid Deep Learning Model for Trash Classification Based on Deep Transfer Learning. J. Electr. Comput. Eng.
2022, 2022, 1–9. [CrossRef]
56. Liu, W.; Ouyang, H.; Liu, Q.; Cai, S.; Wang, C.; Xie, J.; Hu, W. Image recognition for garbage classification based on transfer
learning and model fusion. Math. Probl. Eng. 2022, 2022, 1–12. [CrossRef]
57. Liu, F.; Xu, H.; Qi, M.; Liu, D.; Wang, J.; Kong, J. Depth-wise separable convolution attention module for garbage image
classification. Sustainability 2022, 14, 3099. [CrossRef]
58. Bobulski, J.; Piatkowski, J. PET waste classification method and plastic waste database—WaDaBa. In Advances in Intelligent
Systems and Computing; Springer Verlag: Berlin/Heidelberg, Germany, 2018; Volume 681, pp. 57–64.
59. Wu, Y.; Shen, X.; Liu, Q.; Xiao, F.; Li, C. A garbage detection and classification method based on visual scene understanding in the
home environment. Complexity 2021, 2021, 1–14. [CrossRef]
60. Chen, Y.; Han, W.; Jin, J.; Wang, H.; Xing, Q.; Zhang, Y. Clean our city: An automatic urban garbage classification algorithm using
computer vision and transfer learning technologies. J. Phys. Conf. Ser. 2021, 1994, 012022. [CrossRef]
61. Sun, A.; Xiao, H. ThanosNet: A novel trash classification method using metadata. In Proceedings of the 2020 IEEE International
Conference on Big Data (Big Data), Atlanta, GA, USA, 10–13 December 2020; pp. 1394–1401.
62. Vo, A.H.; Son, L.H.; Vo, M.T.; Le, T. A novel framework for trash classification using deep transfer learning. IEEE Access 2019, 7,
178631–178639. [CrossRef]
63. Hossain, S.; Debnath, B.; Anika, A.; Junaed-Al-Hossain, M.; Biswas, S.; Shahnaz, C. Autonomous trash collector based on object
detection using deep neural network. In Proceedings of the TENCON 2019–2019 IEEE Region 10 Conference (TENCON), Kochi,
India, 17–20 October 2019; pp. 1406–1410.
64. Tiyajamorn, P.; Lorprasertkul, P.; Assabumrungrat, R.; Poomarin, W.; Chancharoen, R. Automatic trash classification using
convolutional neural network machine learning. In Proceedings of the 2019 IEEE International Conference on Cybernetics and
Intelligent Systems (CIS) and IEEE Conference on Robotics, Automation and Mechatronics (RAM), Bangkok, Thailand, 18–20
November 2019; pp. 71–76.
65. Lin, W. YOLO-green: A real-time classification and object detection model optimized for waste management. In Proceedings of
the 2021 IEEE International Conference on Big Data (Big Data), Orlando, FL, USA, 15–18 December 2021; pp. 51–57.
66. Panwar, H.; Gupta, P.K.; Siddiqui, M.K.; Morales-Menendez, R.; Bhardwaj, P.; Sharma, S.; Sarker, I.H. AquaVision: Automating
the detection of waste in water bodies using deep transfer learning. Case Stud. Chem. Environ. Eng. 2020, 2, 100026. [CrossRef]
67. Ma, W.; Wang, X.; Yu, J. A lightweight feature fusion single shot multibox detector for garbage detection. IEEE Access 2020, 8,
188577–188586. [CrossRef]
68. Fulton, M.; Hong, J.; Islam, M.J.; Sattar, J. Robotic detection of marine litter using deep visual detection models. In Proceedings of
the 2019 International Conference on Robotics and Automation (ICRA), Montreal, QC, Canada, 20–24 May 2019; pp. 5752–5758.
69. Sousa, J.; Rebelo, A.; Cardoso, J.S. Automation of waste sorting with deep learning. In Proceedings of the 15th Workshop on
Computer Vision, (WVC), São Bernardo do Campo, Brazil, 9–11 September 2019; pp. 43–48.
70. Gruber, F.; Grählert, W.; Wollmann, P.; Kaskel, S. Classification of black plastics waste using fluorescence imaging and machine
learning. Recycling 2019, 4, 40. [CrossRef]
71. Seredkin, A.V.; Tokarev, M.P.; Plohih, I.A.; Gobyzov, O.A.; Markovich, D.M. Development of a method of detection and
classification of waste objects on a conveyor for a robotic sorting system. J. Phys. Conf. Ser. 2019, 1359, 012127. [CrossRef]
72. Sreelakshmi, K.; Akarsh, S.; Vinayakumar, R.; Soman, K.P. Capsule neural networks and visualization for segregation of plastic
and non-plastic wastes. In Proceedings of the 2019 5th International Conference on Advanced Computing and Communication
Systems (ICACCS), Coimbatore, India, 15–16 March 2019; pp. 631–636.
73. Costa, B.S.; Bernardes, A.C.S.; Pereira, J.V.A.; Zampa, V.H.; Pereira, V.A.; Matos, G.F.; Soares, E.A.; Soares, C.L.; Silva, A.F.B.
Artificial intelligence in automated sorting in trash recycling. In Proceedings of the 2018: Anais do Xv Encontro Nacional de
Inteligência Artificial e Computacional, São Paulo, Brazil, 22–25 October 2018; pp. 198–205.
74. Sakr, G.E.; Mokbel, M.; Darwich, A.; Khneisser, M.N.; Hadi, A. Comparing deep learning and support vector machines for
autonomous waste sorting. In Proceedings of the 2016 International Multidisciplinary Conference on Engineering Technology
(IMCET), Beirut, Lebanon, 2–4 November 2016; pp. 207–212.
75. Li, X.; Su, J.; Wang, H.; Boczkaj, G.; Mahlknecht, J.; Singh, S.V.; Wang, C. Bibliometric analysis of artificial intelligence in
wastewater treatment: Current status, research progress, and future prospects. J. Environ. Chem. Eng. 2024, 12, 113152. [CrossRef]
76. Su, J.; Zhang, F.; Yu, C.; Zhang, Y.; Wang, J.; Wang, C.; Wang, H.; Jiang, H. Machine learning: Next promising trend for
microplastics study. J. Environ. Manag. 2023, 344, 118756. [CrossRef]

Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual
author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to
people or property resulting from any ideas, methods, instructions or products referred to in the content.