RAMAIAH INSTITUTE OF TECHNOLOGY

(Autonomous Institute Affiliated to VTU), Bangalore

Dept. of Electronics and Communication Engineering

MINI PROJECT WORK PHASE– 1

ZEROTH REVIEW

REPORT ON
Underwater Plastic Detection and Congestion Detection
Using Machine Learning

Under the Guidance Of: Report Submitted by:

Shreyansh- (1MS21EC107)
Dr. S. Imaculate Rosaline Tushar Kaushik- (1MS21EC125)
Umang Sonthaliya-
Assistant Professor,
Dept of Electronics and Communication, (1MS21EC127)
RIT Bengaluru Gaurav Singh- (1MS21EC049)
Introduction:
Problem Statement:

Literature Survey:

Methodology:

Tools to be used:

References:
Introduction:
Our environment confronts a myriad of challenges, transcending conventional issues like noise
and air pollution to penetrate the very fabric of life: water. Pollution, the introduction of harmful
substances into our surroundings, manifests in various forms, impacting air, land, water bodies,
and even emitting radiation and loud noises. Among these, water pollution emerges as a grave
concern, casting a shadow over lakes, rivers, and oceans, vital lifelines of our planet. The
relentless influx of plastic waste and hazardous materials dumped into our seas by ships
exacerbates this crisis, with coastal regions disproportionately affected.

Tackling underwater pollution presents formidable challenges due to the harsh conditions of the
aquatic environment. Yet, amidst these challenges, technology emerges as a beacon of hope,
offering innovative solutions to mitigate environmental degradation. Object recognition stands
out as a promising avenue, wherein computers discern specific items in images or videos. This
process, bolstered by advancements in machine learning, particularly deep learning, holds
immense potential for addressing underwater pollution.

Underwater pollutants

Deep learning algorithms, such as Convolutional Neural Networks (CNNs) and You Only Look
Once (YOLO), have revolutionized object recognition, enabling the development of
sophisticated systems capable of detecting and tracking underwater pollutants. By harnessing the
power of these technologies, researchers and environmentalists can gain deeper insights into the
extent and impact of underwater pollution. Moreover, these tools facilitate the implementation of
targeted interventions to mitigate pollution, safeguarding marine ecosystems and preserving the
delicate balance of our oceans.
 Underwater waste materials

Furthermore, the applications of underwater object detection extend beyond pollution mitigation,
encompassing diverse fields such as ocean exploration, resource management, and environmental
conservation. As technology continues to evolve, the potential for innovation in underwater
object detection remains vast, promising to reshape our approach to marine stewardship and
sustainable development.

In essence, the integration of machine learning and object recognition represents a pivotal step
towards addressing the complex challenges of underwater pollution. By leveraging technological
advancements, we can chart a course towards a cleaner, healthier environment, ensuring the
preservation of our oceans for generations to come.

Plastic ingested by a Rainbow Runner Sea shepherd caught in a Plastic Cover

Problem Statement:

During the past few decades, water pollution due to man-made waste has become severe, leading to
contamination of water resources and affecting aquatic life. Despite environmental regulations, trash, litter and
trash objects like plastic are disposed of into the water bodies, affecting the underwater living creatures. This also
creates a huge risk to the very evolution of marine life. We are wrong when we say the impact is only on marine
life it is also to human health as it increases the nitrogen and phosphorus in our waterways and this
overabundance can cause major concerns to human health.

In a recent study, it was observed by a British organization “Verisk Maplecroft” about the growing crisis of trash
objects like plastic and this majorly consisted of plastic. 2.3 billion tons in 2023 to 3.8 billion tons by 2050. Also, the
United States produces 12% of the world’s waste (about 239 million tons), although the country has only 4% of the
global population. India and China contribute to 27% of the total trash objects like plastic.

With the increased usage of Plastic, it has reached the ocean by the source of rivers and many other forms of water
that is flowing through the landfills, and also by dumping from ships, a subsequent contribution from the effluents
from wastewater plants is also evident.
Five major areas of concentration of plastics in the sea have been identified: North Pacific, followed by the Indian
Ocean, the North Atlantic, the South Pacific, the South Atlantic and the Mediterranean Sea. The high concentration
of all forms of micro plastics is in these surface areas.

A paper published on the internet shows that as per the statistics published at Condor Ferries
1. Millions of marine animals die annually due to plastic waste, a significant increase from previous estimates.
2. Over 8.3 billion metric tons of plastic debris are present in oceans, with around 5.25 trillion plastic pieces.
3. Approximately 80% of waste debris sinks into the ocean, while 10% remains suspended and 10% washes ashore.

There are trash objects like plastic patches in the great pacific, which is also known as pacific trash vortex and it
covers around 617K miles between the region of Hawaii and California. And this is still a small part of the entire
marine pollution.
Our coasts and the coastlines are the regions that have high concentrations of plastic, especially in regions which
has high coastal populations, inadequate waste management systems, intensive fishing industries, or high tourism.

Ideally, every region in the world should have a proper system for trash objects like plastic collection and recycling
of waste and the control of landfill waste. Since this is being implemented in some areas, plastics and generate
need to be collected. This remains a big challenge as waste and plastic is dispersed in many areas of the planet.

Aquatic life affected Sea turtles ingested plastic

Literature Survey:

Manual Survey:
 Manual surveys entail divers visually inspecting underwater environments to identify and collect plastic
debris.
 This method relies on human observation and physical collection of plastic waste, making it labor-intensive
and time-consuming.
 Manual surveys are effective for small-scale assessments or localized areas where detailed inspection is
required.
 However, the coverage area of manual surveys is limited due to the logistical constraints of diver
operations.
 Additionally, manual surveys may be impractical or hazardous in certain underwater environments, such as
deep-sea locations or areas with strong currents.
 Despite its limitations, manual surveys provide valuable data for understanding the distribution and
abundance of plastic pollution in specific underwater habitats.

Remote Sensing:
 Remote sensing, including satellite imagery and aerial drones, is utilized for detecting and monitoring
plastic pollution in oceans and coastal areas.
 Satellite-based sensors capture vast ocean surface areas, facilitating detection of plastic accumulation over
large spatial scales.
 Aerial drones equipped with high-resolution cameras provide detailed images of coastal regions, aiding in
the identification of plastic debris along shorelines and near the water's surface.
 Drones maneuver close to the water's surface, enabling precise detection and mapping of plastic debris in
coastal areas.
 High-resolution drone imagery helps assess the extent and impact of plastic pollution on coastal
ecosystems.
 Remote sensing data analysis reveals patterns and trends in plastic pollution distribution, informing targeted
mitigation strategies and conservation efforts for marine environments.

Underwater Robotics:
 Underwater robotics, encompassing ROVs and AUVs, revolutionize underwater surveys and monitoring for
plastic debris.
 Equipped with sensors, cameras, and manipulator arms, these vehicles can detect and collect plastic debris
from underwater environments.
 ROVs are operated by human operators, allowing real-time navigation and manipulation, while AUVs
operate autonomously.
 AUVs are suitable for long-duration missions in remote or hazardous underwater locations due to their
autonomous operation.
 Both types of underwater robotics can efficiently map underwater terrain, identify plastic debris hotspots,
and conduct targeted surveys.
 Sophisticated sensor suites, including sonar and environmental sensors, enable underwater robotics to
navigate and explore challenging underwater environments.
 Underwater robotics play a crucial role in understanding the extent and impact of plastic pollution on
marine ecosystems.
By leveraging the capabilities of underwater robotics, researchers and environmental agencies can enhance their
efforts to mitigate the adverse effects of plastic pollution on underwater habitats and biodiversity.
Sensors:
Acoustic sensors:
 Acoustic sensors, such as hydrophones, are designed to detect sound waves underwater. When objects,
including plastic debris, interact with water, they produce acoustic signals that can be detected by these
sensors. Acoustic signals generated by plastic debris may vary depending on factors such as size, shape, and
material composition. By analyzing the characteristics of these acoustic signals, researchers can distinguish
plastic debris from other underwater objects and background noise. Acoustic sensors can be deployed on
underwater platforms, such as buoys or autonomous underwater vehicles (AUVs), to conduct real-time
monitoring of plastic pollution in oceans and coastal areas.

Optical Sensors:
 Optical sensors utilize light to detect and characterize underwater objects. These sensors can capture images
or video footage of underwater environments, allowing researchers to visually identify and classify plastic
debris. Optical sensors can be equipped with cameras, spectroscopes, or other imaging devices to capture
detailed information about the size, shape, color, and texture of plastic debris. By analyzing optical data,
researchers can estimate the abundance, distribution, and spatial patterns of plastic pollution in underwater
ecosystems. Optical sensors are commonly integrated into underwater vehicles, such as remotely operated
vehicles (ROVs) or underwater drones, for systematic surveys and monitoring of plastic debris in marine
environments.

Significance of Sensors in Underwater Plastic Detection:

 Acoustic and optical sensors play a crucial role in underwater plastic detection by providing complementary
data streams for identifying and characterizing plastic debris. Acoustic sensors can detect plastic debris
based on its acoustic signature, while optical sensors enable visual confirmation and classification of plastic
debris. By combining data from both sensor types, researchers can improve the accuracy and reliability of
underwater plastic detection systems. Furthermore, acoustic and optical sensors can be deployed on
autonomous or remotely operated platforms, allowing for efficient and scalable monitoring of plastic
pollution across large underwater areas. Overall, the integration of acoustic and optical sensor technology
enhances our ability to understand and mitigate the impacts of plastic pollution on marine ecosystems.

Machine Learning:
Machine learning algorithms are used to process large volumes of data and identify patterns associated with plastic
pollution. These algorithms learn from labeled datasets to recognize characteristics of plastic debris, such as shape,
texture, and color, enabling more efficient and accurate detection methods. Several ML algorithms and techniques
are commonly used for this purpose:

 Convolutional Neural Networks (CNNs): CNNs are deep learning algorithms specifically designed for
processing visual data, such as images. They consist of multiple layers of convolutional filters that extract
features from input images, followed by pooling layers that reduce dimensionality. CNNs have been
successfully applied to image classification tasks, including the detection of objects underwater. Libraries
commonly used for implementing CNNs include TensorFlow, Keras, and PyTorch.

 Support Vector Machines (SVM): SVM is a supervised learning algorithm used for classification tasks. It
works by finding the optimal hyperplane that separates different classes in feature space. SVM has been
used for object detection and classification, including the identification of plastic debris underwater.
Libraries such as scikit-learn provide implementations of SVM for classification tasks.

 Random Forests (RF): RF is an ensemble learning algorithm that combines multiple decision trees to make
predictions. Each tree in the forest independently classifies input data, and the final prediction is determined
by a majority vote. RF is often used for classification tasks, including the detection of underwater objects.
Libraries such as scikit-learn offer RF implementations for classification tasks.
  Deep Learning Architectures (Faster R-CNN): Deep learning architectures designed specifically for object
detection, such as Faster R-CNN,is also used for detecting plastic debris underwater. These architectures
can localize and classify objects in images with which appreciable accuracy. Implementations of these
architectures are available in frameworks like TensorFlow Object Detection API and PyTorch torchvision.

 Unsupervised Learning Techniques: Unsupervised learning techniques, such as clustering algorithms (e.g.,
k-means), can be used for exploratory analysis of underwater imagery to identify clusters of potential
plastic debris. While these techniques do not require labeled data, they can provide insights into the
distribution and patterns of plastic pollution underwater.

Libraries commonly used for implementing machine learning algorithms and conducting data analysis include
TensorFlow, Keras, PyTorch, scikit-learn, and OpenCV (for image processing tasks). These libraries provide
extensive functionalities for training ML models, processing data, and evaluating model performance.
Methodology:
Our approach to solving the problem is defined by employing a comprehensive comparison between the Fast R-
CNN and YOLO (You Only Look Once) object detection methodologies.

Architecture:
 Fast R-CNN: Fast R-CNN is a region-based convolutional neural network (CNN) architecture. It proposes
regions of interest (RoIs) using a region proposal network (RPN) and then extracts features from these RoIs
for classification and bounding box regression.

 YOLO: YOLO is a single-stage object detection architecture that directly predicts bounding boxes and class
probabilities from a single neural network. It divides the input image into a grid and predicts bounding
boxes and class probabilities for each grid cell simultaneously.
Speed:

 Fast R-CNN: Fast R-CNN is slower than YOLO because it requires multiple forward passes through the
network for region proposal, feature extraction, and classification.

 YOLO: YOLO is faster than Fast R-CNN because it performs object detection in a single pass through the
network, making it more suitable for real-time applications.
Accuracy:

 Fast R-CNN: Fast R-CNN tends to have higher accuracy than YOLO, especially for small objects and in
scenarios where precise localization is critical. It achieves this by performing multi-scale feature extraction
and using RoI pooling for accurate bounding box alignment.

 YOLO: YOLO sacrifices some accuracy for speed. While it can achieve good performance, especially on
large objects, it may struggle with small objects and fine-grained details due to its single-stage architecture.

Training Complexity:

 Fast R-CNN: Training Fast R-CNN involves multiple stages, including training the region proposal
network (RPN) and the subsequent fine-tuning of the CNN backbone. This process can be more complex
and computationally intensive.
 YOLO: YOLO has a simpler training process since it involves training a single neural network end-to-end
for object detection. This simplicity makes YOLO easier to train and deploy.

Robustness to Object Size:

 Fast R-CNN: Fast R-CNN is more robust to variations in object size and aspect ratio due to its multi-scale
feature extraction and RoI pooling mechanism.

 YOLO: YOLO may struggle with small objects or objects with irregular shapes, as it processes the entire
image at once and may not capture fine-grained details effectively.

We can conclude that, Fast R-CNN offers higher accuracy and better localization capabilities but
is slower compared to YOLO. YOLO, on the other hand, sacrifices some accuracy for speed and
simplicity, making it more suitable for real-time applications where speed is paramount. The
choice between the two methods depends on the specific requirements of the object detection
task, including accuracy, speed, and computational resources available.
Tools to be used:
 Python: Python is a versatile programming language widely used in machine learning projects. It offers
numerous libraries and frameworks specifically designed for machine learning tasks.

 TensorFlow: TensorFlow is an open-source machine learning framework developed by Google. It

provides comprehensive support for building and training machine learning models, including deep learning
architectures.

 PyTorch: PyTorch is another popular open-source machine learning framework known for its flexibility
and ease of use. It offers dynamic computation graphs and extensive support for deep learning model
development.

 scikit-learn: scikit-learn is a powerful Python library for machine learning tasks such as classification,
regression, clustering, and dimensionality reduction. It provides a wide range of algorithms and tools for
building machine learning pipelines.

 OpenCV: OpenCV (Open-Source Computer Vision Library) is a library of programming functions

mainly aimed at real-time computer vision tasks. It provides tools for image processing, feature detection,
and object recognition, which are essential for preprocessing and analyzing underwater images.

 Keras: Keras is a high-level neural networks API written in Python, capable of running on top of
TensorFlow, Theano, or Microsoft Cognitive Toolkit (CNTK). It offers a user-friendly interface for
building and training deep learning models.

 Matplotlib and Seaborn: Matplotlib and Seaborn are Python libraries used for data visualization.
They provide functions for creating various types of plots and charts, allowing for better understanding and
analysis of data.

 Pandas: Pandas is a Python library used for data manipulation and analysis. It provides data structures
and functions for efficiently handling structured data, such as loading datasets, performing data cleaning,
and conducting exploratory data analysis.

 NumPy: NumPy is a fundamental package for scientific computing with Python. It provides support for
large, multi-dimensional arrays and matrices, along with a collection of mathematical functions to operate
on these arrays.

 Jupyter Notebooks: Jupyter Notebooks are interactive computing environments that allow for the
creation and sharing of documents containing live code, equations, visualizations, and narrative text. They
are widely used for prototyping and experimenting with machine learning models.
References:
Huang, Yonghui, et al. "Research on evaluation method of underwater image quality and performance of
underwater structure defect detection model." Engineering Structures 306 (2024): 117797.

Xu, Shubo, et al. "A systematic review and analysis of deep learning-based underwater object detection."
Neurocomputing 527 (2023): 204-232.

Kalaiarasi, G., et al. "A Deep Learning Approach to Detecting Objects in Underwater Images." Cybernetics and
Systems (2023).

Goxhaj, Ornelta, et al. "Underwater image detection for cleaning purposes; techniques used for detection based on
machine learning." Acta Marisiensis. Seria Technologica 19.1 (2022): 28-35.

Stafford, Richard, and Peter JS Jones. "Viewpoint–Ocean plastic pollution: A convenient but distracting truth?."
Marine policy 103 (2019): 187-191.

Scotti, Gianfranco, et al. "Seafloor litter along the Italian coastal zone: an integrated approach to identify sources of
marine litter." Waste Management 124 (2021): 203-212.