IRJET-Multiclass Classification Method Based On Deep Learning For Leaf Identification For Farmers

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 05 Issue: 06 | June-2018 www.irjet.net p-ISSN: 2395-0072
© 2018, IRJET | Impact Factor value: 6.171 | ISO 9001:2008 Certified Journal | Page 1126
Multiclass Classification Method Based On Deep Learning For Leaf
Identification For Farmers
Sanjay Sunkad1, Aakash Bagale2, Srujan Rumale3, Zabiulla Shariff4
1,2,3,4Department of Computer Science and Engineering, Maharaja Institute of Technology Mysore
Approved by AICTE, New Delhi, Affiliated to VTU, Belgaum, Recognized by Government of Karnataka
Mysore INDIA
---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract -There are estimated to be nearly half a million
species of plant in the world. Classification of species has
been historically problematic and often results in duplicate
identifications. Plant identificationbasedonleafisbecoming
one of the most interesting and a popular trend. Each leaf
carries unique information that can be used in the
identification of plants. In the identification of plants based
on leaf, the leaf images need to be preprocessed accordingly
to extract the various critical features.
As the upcoming ground-breaking performance of neural
networks in the field of classification and identification of
objects is being carried out where success is being achieved,
the new neural network known as Convolutional Neural
Network(CNN) has made theidentificationandclassification
of objects more reliable with the performance and as well as
the computations. Like any other classifier, the
Convolutional Neural Network (Model) is trained with
images and its specific labels. Having few layered Neural
Network; we will get patterns (features) in given object
(image). This trained model is used to classify new input
images.
As supervised deep learning is used it makes it produces
accurate results and makes it easier for the user to classify
leaf data with much higher accuracy. CNN’s are great at
images, and have promise for text, simply because they are
capable of identifying patterns in huge number of
homogeneous features (pixels in 2D or characters in 1D).
They are designed to handle large amount of image data.
Hence by using Convolutional Neural Network, the accuracy
is drastically improved.
Key Words: HSV – Hue Saturation Value, CNN –
Convolutional Neural Network, ReLU – Rectified
Linear Unit
1. INTRODUCTION
Plant identification based on leaf is becoming one of the
most interesting and a popular trend. Each leaf carries
unique information that can be used in the identification of
plants. In the identification of plants based on leaf, the leaf
images need to be pre-processed accordingly to extract the
various critical features.
The proposed approach consists of three phases that are
pre-processing, feature extraction, andclassificationphases.
Since most types of plants have unique leaves, so the
classification approach presented in this research depends
on plants leave. Leaves are different from each other by
characteristics such as the shape, colour, texture and the
margin. Dataset are used for both training and testing.
Neural networks are used here. Convolutional Neural
Networks are very similar to ordinaryNeural Networksthey
are made up of neurons that have learnable weights and
biases. Each neuron receives some inputs, performs a dot
product and optionally follows it with a non-linearity.
The whole network still expresses a single differentiable
score function: from the raw image pixelsonone endtoclass
scores at the other. And they still have a loss function (e.g.
SVM/ Softmax) on the last (fully-connected)layerandall the
tips/tricks we developed for learning regular Neural
Networks still apply.
1.1 RELATED WORK
Talking about the existingsystemtherearemanytraditional
machine learning methods like Knearest neighbour (KNN),
Support vector machine (SVM) result in lower accuracy in
prediction leading to miss classification of input image.
Random forests, KNN, SVM have strengths in other type of
problem: mainly in the space where there are relatively few
(tens or hundreds) of heterogeneous features. CNN’s are
great at images, and have promise for text, simply because
they are capable of identifying patterns in huge number of
homogeneous features (pixels in 2D or characters in 1D).
They are designed to handle large amount of image data.
Hence by using Convolutional Neural Network, the accuracy
is drastically improved.
Traditional Image processing uses linear classifier like
support vector machine or K Nearest Neighbours (KNN)
where it fails to classify new data points into its specific
class. If such classifier is used to classify bulk images it could
lead to reduction of theirqualitybecauseof misclassification.

1.2 METHEDOLOGY
Figure 1 Flow of process
A. Pre-processing The input image is first converted into
HSV color space model and then using an inrange function
which extracts majority of the green pixels present in the
image. Further a bounding rect function is used to crop the
image where the unwanted region is eliminated. The
cropped image that is region of interest (ROI),whichisgiven
as input to the next step for identification.
B. Feature Extraction and classification CNN model is
used for feature extraction where the data is analyzed in the
following manner where the CNN model automatically
extracts thousands of features. The image is passed through
a stack of convolutional layers. Response normalization
layers follow the first and second convolutional layers.
Spatial pooling is carried out by four maxpooling layers. The
ReLU non-linearity is applied to the output of every
convolutional and fully-connected layer.
Figure 2 Inception V3 model
K-means clustering:
Equation 1
Back propagation:
Figure 3

2 WORKING
Figure 4
Module 1: Leaf capture using OpenCV:
Leaf is captured using OpenCV. The image of the leaf can be
captured by using 3 methods Default camera: The laptop
built in camera can be used to capture the image of the
individual. IP camera can make use of mobile device and
capture HD images of leaf to get better accuracy and results.
Web camera can also be accessed to get the image of leaf.
Module 2: leaf Recognition using Convolutional Neural
Network:
Leaf recognition has always been a challenging task from
traditional image processing technique to modern image
processing. To obtain higher accuracy and better prediction
of the recognized Convolutional neural network is used
where Artificial Intelligence Model (CNN) needs to be
integrated. This involves the following steps. Collecting the
dataset is an important work where we need to collect
images of leaves. If we have 2 people, then we need a dataset
of at least 20-25 images. More the data bettertheaccuracyin
modern AI. Training the Model: We need to train the new
leaves (20-25 images) and teach our model which the
particular leaf is for example, we load 25 imagesofleaf1and
train the model and 20 images of leaf 2 and train the model.
Leaf Recognition: Finally, we should be able to detect the
leaves that we have been trained with good prediction
accuracy. Retraining the model: If the model has lower
accuracy then we need to retrain the model and start from
step1 for better accuracy. We need to make use of Tensor
flow (Google’s Open Source library) for face recognition
where python is used as a base language.
Module 3: leaf disease recognition:
The diseased leaf needs to be identified. The disease may be
categorized into many forms. We first need to identify the
disease of the leaf by giving the image of the leaf. Pre-
training the model is necessary as it was done in module 2.
The diseased leaf is categorized into its respective category.
Creating an UI using C# where an UI is integrated with some
buttons, picture boxes, timers as per coding requirement.
Module 4: Connection leaf recognitioncode/leafdisease
recognition code with that of OpenCV C++ library:
The OpenCV C++ code needs to be connected withthatofthe
OpenCV code to make it simpler. Module 5: Display the final
output on the UI Final output has to be displayed on the
output with no glitch and error free.
3. CONCLUSIONS
We have successfully identified 13 categories of plantleaves
and 4 categories of diseased leaves. The accuracy of our
model is about 95.25%. Model has a huge influence on the
lighting conditions while capturing from IP camera or
webcam.
ACKNOWLEDGEMENT
We would like to show our gratitude towards our principal,
Dr. B G Naresh Kumar for his great support.
We extend our warm wishes to our HOD, Prof. Deepu. R
and our guide Asst. Prof Sushma Koushik N. Special and
sincere thanks to our beloved co-ordinator, Prof. Hemanth
S R for his inspiration and encouragement.
REFERENCES
[1] Siang Thye Hang*, Masaki Aono, “Open World Plant
Image Identification Based on Convolutional Neural
Network”, IEEE 2016.
[2] Ajla KULAGLIC and B . BerkUSTUNDA, “Neural
Network with Wavelet Preprocessing for Wheat
Growth stage”, IEEE 2016.

[3] Jobin Francis, Anto Sahaya Das D and Anoop B K,
“Identification of Leaf Disease in Pepper Plants
Using Soft Computing Techniques” , IEEE 2016.
[4] Yuanyuan Liu, Deng Wenzhou, Fan Tang, Yipping
Meng and Weiming Dong, “Flower Classification via
Convolutional Neural Network”, IEEE 2016
[5] Adil Salman, Ashish Semwal, V. M Thakkar and
Upenda Bhatt,“Leaf Classification and Identifiaction
Using Canny Edge Detector and SVM Classifer ”,
IEEE 2016.
[6] Suhas S Athani and CH Tejaswini, “Support Vector
Machine-Based Classification Scheme of Maize Crop
“, IEEE 2017.
[7] Anjali Yadav, Jesus B. Alonso , Carlos M. Travies
,Malay Kishore Dutta “ Automatic Identification of
Botanical Samples of Leaves using Computer Vision
“, IEEE 2017.
[8] Monzural Islam, Anh Dinh, Khan Wahid and Pankaj
Bhowmik, ” Detection of Potato Diseases Using
Image Segmentation and Multiclass Support Vector
IEEE 2017.
BIOGRAPHIES
Sanjay Sunkad
Dept of CSE
MIT Mysore
Aakash Bagale
Dept of CSE
MIT Mysore
Srujan Rumale
Dept of CSE
MIT Mysore
Zabiulla Shariff
Dept of CSE
MIT Mysore

IRJET-Multiclass Classification Method Based On Deep Learning For Leaf Identification For Farmers

More Related Content

What's hot (20)

Similar to IRJET-Multiclass Classification Method Based On Deep Learning For Leaf Identification For Farmers (20)

More from IRJET Journal (20)

Recently uploaded (20)

IRJET-Multiclass Classification Method Based On Deep Learning For Leaf Identification For Farmers