LOVELY PROFESSIONAL UNIVERSITY

PHAGWARA, PUNJAB, INDIA

FLOWER RECOGNITION WITH MACHINE LEARNING

1Uday Pratap Pandey
[email protected]
School of Computer Application

ABSTRACT:
Recognizing different types of flowers has been a tough problem in computer vision for a long time. Previous attempts to solve
it using traditional machine learning methods didn't work well because they either had small datasets or were too simple. In this
paper, we introduce a new way to classify flower species using deep learning. We use a type of neural network called a
convolutional neural network (CNN) to automatically learn the unique features of each flower species. Instead of manually
designing these features, our model learns them on its own. We trained our model on a dataset containing five different types of
flowers, allowing it to learn what makes each flower species distinct. Then, we tested our model on new data to see how well it
could identify flower species. Our experiments show that our approach performs much better than traditional methods, proving
the power of deep learning in solving complex visual recognition tasks like flower classification.

Keywords: Image Classification, Machine Learning, Convolution Neural Network(CNN).

INTRODUCTION:
Flower recognition using Convolutional Neural Networks PROBLEM DEFINITION
(CNN) is a fascinating area of research and practical
Similar initiatives have lately been created for identifying
application in the field of computer vision. CNNs are a type
both plants and flowers through their leaves. Since there
of deep learning model that are particularly well suited for
are numerous variations in flower shape, color, and texture,
image classification tasks, making them an ideal choice for
the creators' biggest issue is determining the right feature
recognizing different types of flowers based on their visual
extraction criteria related to plants and flowers. It was
characteristics. The goal of flower recognition using CNN
noted that most of the systems employed in these
is to train a model that can accurately identify different
experiments concentrated on the computational models
species of flowers based on their unique features, such as
logic involved in input images. As a result, the semantic gap
color, shape, and texture. This is achieved by feeding the
that results from the disparity between how humans
CNN model with a large dataset of labelled images of
perceive an image and how it is represented digitally was
different flower species, allowing the model to learn to
identified as the main problem.
recognize patterns and features specific to each flower
type. Flower recognition has numerous applications, from OBJECTIVE
aiding botanists in their research to assisting in the The most of the methods used today result in unproductive
development of smart agricultural systems. By accurately results, such as predicting only a few projected blooms with
identifying different types of flowers, we can gain a better a high degree of likelihood. Hence, our objective is to create
understanding of the natural world and help to protect it accurate and simple for common people to identify a
and preserve our natural resources. In this project, we will flower they have been seeing. We develop a powerful model
explore the process of building a flower recognition for the categorization of floral photos in the proposed
system using CNN, including data pre-processing, model system using convolutional neural networks. Also, it
architecture, and training and evaluation techniques. emphasizes the primary functions of the recognized flower,
Through this work, we can gain a deeper understanding of improving system performance.
the potential of CNNs and their role in advancing the field
of computer vision.
1
Page
 Why use Neural Network Segmentation

Neural network with their remarkable ability to derive Segmentation divides an image once more into its individual
meaning from complicated or imprecise data can be used to objects or regions. Discontinuity and similarity, which are two
extract patterns and detect trends that are too complex to be qualities, are used to categorize picture segmentation
noticed by either humans or other computer techniques. techniques. Due to this attribute, there are two types of
A trained neural network can be thought of as an “expert” in picture segmentations one is region-based segmentation and
the category of information it has been given to analyze. This another one is edge-based segmentation.
expert can then be used to provide projections given new
Feature Extraction
situations of interest and answer “what if” questions. Other
Advantages Include: Features define classification functions and contain

Adaptive Learning: An ability to learn how to do tasks based information about the target.

on the data given for training or initial experience. Classification/Regression

Self-Organization: A CNN can create its own organization or
The labelling of photographs into one of the several specified
representation of the information it receives during learning
categories is referred to as image classification. This section
time.
has covered primary classification approaches along with
their primary.
METHODOLOGY
This is an illustration of the method or ability for flower RESULTS AND DISCUSSION
prediction based on class type. The fact that the result There are three components to the general difficulty of floral
from one phase serves as the source for the following may species identification. Initially, Over Feat CNN network is used
be seen as a waterfall model. This paradigm's stages each to extract features from the photos in the training data and
have a clear purpose to serve and an explanation for why then the features are indexed into an HDF5 file format.
they are there. So let's take a quick look at the events in the Second, multiple machine learning techniques are used to
sequence they occurred. train the network.

In order to assess the system's precision, random sample

images are lastly provided to the networks for label
prediction.

The dataset was gathered from numerous online domains. It

is a dataset of images. contains pictures of a variety of flowers,
including Hibiscus, Lotus, Marigold, Rose, Periwinkle, Touch
Me Not, Zennia, and Jasmine. The Train directories and a
Validate directory may be found in the dataset. 80% of the
photos are used for training and 20% are used for validation.
On a total of 2512 photos from 8 classes, the model was
evaluated. We use the 2010 documents for training and the
502 files for testing out of these. On average, the algorithm
Image Acquisition
successfully recognizes flower species with 80% accuracy.
The general aim of Image Acquisition is load the data from
the dataset. The data set is collected from various websites
in the internet.

Image Processing

According to preprocessing, different photos of the same

type of tissue may have a variable scale of signal intensities.
2
Page
CONSLUSION 6. Parvathy S N, N Vrinda Rao, Shahistha Bai S, Naeema Nazer,
In conclusion, flower recognition using CNN is an exciting Prof. Anju J, "Flower Recognition System Using CNN," 2020.
and rapidly developing area of research in computer vision.
7. Ferhat Bozkurt, "A Study on CNN Based Transfer Learning
With the growing availability of large and diverse datasets,
for Recognition of Flower Species," 2021.
as well as the increasing power and sophistication of deep
8. I. Gogul, V. Sathiesh Kumar, "Flower Species Recognition
learning models, it is now possible to build highly accurate
System using Convolution Neural Networks and Transfer
and reliable systems for automatically identifying different
Learning," 2017.
types of flowers based on their visual features. The ability to
recognize flowers has numerous practical applications, from 9. Rabindra Patel, Chandra Sekhar Panda, "A Review on
aiding botanists in their research to developing smart Flower Image Recognition," 2019.
agricultural systems and monitoring the health of
10. Neda Alipour, Omid Tarkhaneh, Mohammad Awrangjeb,
ecosystems. By accurately identifying different types of
"Flower Image Classification Using Deep Convolutional
flowers, we can gain a deeper understanding of the natural
Neural Network."
world and help to protect and preserve our natural
resources. However, the challenges of flower recognition,
including variability in appearance, limited data availability,
and computational requirements, must be carefully
addressed through rigorous data collection, pre-processing,
model selection, and evaluation. With the right approach and
a commitment to excellence, we can continue to push the
boundaries of what is possible in flower recognition using
CNN and advance the field of computer vision as a whole.

REFERENCES
1. Mastura Hanafiah, Mohd Azraei Adnan, Shuzlina Abdul-
Rahman, Sofianita Mutalib, Ariff Md Ab Malik, Mohd Razif
Shamsuddin, "Flower Recognition Using Deep Convolutional
Neural Networks," 2021.

2. Archana L. Lakesar, "A Review on Flower Classification

Using Neural Network Classifier," 2017.

3. Thi Thanh Nhan Nguyen, Van Tuan Le, Thi Lan Le, Hai Vu
Natapon Pantuwong, Yasushi Yagi, "Flower Species
Identification Using Deep Convolutional Neural Networks,"
2016.

4. Rekha A Shidnekoppa, Deepti Aralikatti, Vinuta

Bangarshettar, Ashwini Koti, Shree Halbhavi, "Automated
Flower Species Detection and Recognition using Neural
Network," 2020.

5. Hazem Hiary, Heba Saadeh, Maha Saadeh, Mohammad

Yaqub, "Flower classification using deep convolutional
neural networks," 2018.
3
Page