Content uploaded by Vijaya Adluri
Author content
All content in this area was uploaded by Vijaya Adluri on Apr 10, 2024
Content may be subject to copyright.
Chapter 13
Potato Leaf Disease Detection
and Classification Using VGG16
Vijaya Lakshmi Adluri , Pranavi Sai Reddy, Sonalika Palthya,
Saiteja Jampani, and Sai Priya Gandhari
1 Introduction
Potato (Solanum Tuberosum) is an essential and widely cultivated crop that belongs
to the Solanaceae family. It is a starchy tuberous crop that serves as a staple food
for millions of people around the world. Because India’s economy relies mainly
on agriculture with a diverse crop portfolio, potatoes are a significant part of their
society. India is now the second leading producer of potatoes globally, producing over
43 million tons in 2018. Nonetheless, despite the increasing global demand and the
opportunity for India to export more, potato production expansion arises as the most
important factor. Unfortunately, this is not easy, as significant diseases such as early
blight and late blight have caused potato exports and yields to decline in recent years.
As a result, local farmers are facing various challenges. Early and late blights are
common potato diseases that affect farmers in the region. The symptoms of late blight
disease can be observed as blistered spots on leaves that eventually decay and dry up,
while early blight disease manifests as small, black lesions. The use of convolutional
neural networks (CNN) and the VGG16 algorithm will be of great benefit to farmers,
V. L. A d l ur i (B) · P. S. Reddy · S. Palthya · S. Jampani · S. P. Gandhari
Department of Computer Science and Engineering, B V Raju Institute of Technology, Narsapur,
Medak 502313, India
e-mail: Vijaya.a@bvrit.ac.in
P. S. Reddy
e-mail: 21211A05r5@bvrit.ac.in
S. Palthya
e-mail: 21211A05l3@bvrit.ac.in
S. Jampani
e-mail: 21211A05r8@bvrit.ac.in
S. P. Gandhari
e-mail: 22215A0521@bvrit.ac.in
© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2024
M. S. Uddin and J. C. Bansal (eds.), Proceedings of International Joint Conference
on Advances in Computational Intelligence, Algorithms for Intelligent Systems,
https://doi.org/10.1007/978-981-97-0180-3_13
147
148 V. L. Adluri et al.
as it can help distinguish between these diseases and potato leaves. There are three
categories of processed images available, which include healthy leaves, early stage
late blight, and fully developed late blight. The pictures are divided into two groups,
one for training and the other for validation, to help farmers identify and treat these
diseases using appropriate fertilizers.
Recently, there has been a growing interest in using advanced technologies such as
computer vision and deep learning in identifying and classifying plant leaf diseases.
Convolutional neural networks (CNNs) have shown great potential in tasks such as
detecting and diagnosing plant diseases using image identification. Several studies
have been conducted to explore new perspectives on the characterization of plant
diseases using deep learning techniques like CNNs, fuzzy image enhancement
methods, and VGG16 algorithm. These approaches aim to create accurate and reli-
able detection methods for plant diseases using computer-assisted technologies. In
this context, developing a CNN-based system for potato leaf disease detection and
classification holds immense potential to improve disease management and increase
crop productivity. By automating this process, farmers and agricultural experts can
quickly identify diseased plants, implement targeted interventions, and minimize
crop losses.
2 Literature Review
Dataset: The model is trained using a dataset, which is a collection of data. Image
information: The project’s dataset is the Kaggle PlantVillage dataset. Deep learning
encompasses a variety of algorithms, including convolutional neural networks
(CNNs), which are commonly used for tasks such as image classification and object
recognition. CNNs are particularly suited for analyzing visual data using artificial
neural networks, and they employ a grid-like structure that allows them to process
data in a more efficient manner. Through the use of CNNs, deep learning has shown
promise in a range of applications, from medical image analysis to computer vision
and object detection tasks, despite the fact that there are various types of neural
networks.
The author has detected the diseased leaf and area of the leaf affected using CNN
model and has achieved an accuracy of 95.86% [1]. The author has presented a system
that uses deep learning to classify two types of diseases in potato plants based on leaf
conditions, using the GoogleNet, ResNet50, and the VGG16 convolutional neural
network architecture model to create an accurate classification system. The model
has achieved 97% accuracy for the first 40 CNN epochs, indicating the feasibility of
the deep neural network approach [2]. In another analysis, the author has performed
picture division with more than 2000 pictures of healthy and unhealthy potato’s
leaf, which are collected from Kaggle and a few preprepared models are utilized
for acknowledgment and classification of healthy and diseased leaves. The proposed
model has predicted with an accuracy of 91.41% in testing with 30% test data and
13 Potato Leaf Disease Detection and Classification Using VGG16 149
70% train data [3]. A novel plant leaf disease classification with image processing
technique is introduced, and best results were obtained with very little computational
effort [4].
The other approach proposed has used CNN with a maximum of 50 epochs for
the training model, with a batch size of 32. The model was observed to outperform
with 99.91% accuracy during training, 99.80% accuracy during validation, and 100%
accuracy in the testing phase [5]. The CNN model was used and deployed on the
Streamlit platform to classify the leaf disease [6]. In another research, the author
has modified the existing MobileNet-V2 architecture using the fine-tuning approach
and the proposed approach separately attained an average accuracy and specificity of
97.33 and 98.39% on the locally collected image dataset [7]. In the papers [8, 9], the
author has performed feature extraction with K-means clustering and classification
with ANN and has outperformed the existing models.
A new work is done in detecting plant diseases using OpenCV in [10]. In [11], the
author has selected VGG16, VGG19, and ResNet50 network models. Using VGG16,
the author has achieved 97% accuracy which is the best-provided results among
three networks. In [12], the author has performed disease classification with image
processing and random forest technique and the average accuracy achieved was 93%.
In [13], the author has performed the detection of plant diseases by analyzing the
texture of leaf using ANN classifier. In [14], the author has presented a system to
classify the four types of diseases in potato plants based on leaf conditions by using
VGG16 and VGG19 convolutional neural network architecture models to obtain an
accurate classification system. This experiment has achieved an average accuracy
of 91%. In [15], the author has used the pretrained model VGG 19 to predict the
diseases in potato crops using leaf images and has achieved an accuracy of 97.8%.
3 Proposed System
In this work, VGG16 is used to distinguish between diseased and unimpaired leaves
as well as to spot defects in affected plant leaves. Using pictures to train the model
and the input leaf to decide output, the CNN model is made to work with both healthy
and diseased leaves. In the same way that human eyes and brains do, VGG16 can
identify as many images as it can at a time. We are incredibly adept at observing,
comprehending, picking up on, and inferring from a scene because it occurs in our
brains so quickly. CNNs are necessary to help a computer comprehend and interpret
a scene. The proposed work will be able to detect the diseased area on the leaf and
classify diseases into late blight, early blight, or a healthy leaf.
The architecture used is VGG16, and the dataset used is the PlantVillage dataset,
which consists of numerous images of different leaves. The dataset is then trained
using VGG16. The image needs to be processed in order to find the sickness on the
leaves. Image processing is an important field that has many practical applications,
including in computer vision and medical imaging. The process begins with image
acquisition, which is also known as pretreatment. Once the image is acquired, a
150 V. L. Adluri et al.
variety of techniques can be used to extract specific features or enhance its appear-
ance, including image enhancement and restoration. Difficult characteristic of image
processing is segmentation, which involves splitting an image into its discrete objects
or components. Despite its challenges, image processing technology continues to
advance, with new research focusing on areas such as artificial intelligence and
machine learning. These are the steps involved in image processing. Then the final
model is built and executed.
3.1 Architecture
This architecture in Fig. 1 illustrates that it gathers a dataset of potato leaf images,
including both healthy leaves and different diseased leaves. Preprocess the images
by resizing them to a constant size, then split the dataset into two different sets
namely training set and validation set. Transfer learning mechanism helps to improve
the performance and its working mechanism is shown in Fig. 2. Here, the VGG16
architecture has feature extraction by removing the final fully connected layers and
the adding custom layers for the disease classification, and the working model of
VGG 16 is shown in Fig. 3. Then it tests the trained model on the validation set to
access its performance metrics and its generalization ability.
3.2 Transfer Learning
Transfer learning is a machine learning technique, learning new task is dependent
on prior knowledge.
3.3 VGG 16 Architecture
The VGG16 architecture is a convolutional neural network architecture that was
developed by Visual Geometry Group and its input to the VGG16 network is an
image of fixed size and it contains convolutional blocks which consist of five groups
of convolutional layers, each followed by a max pooling layer. After the convolutional
and pooling layers, the network ends with fully connected layers. For the disease
detection here, we remove the last fully connected layers and replace them with
custom classification layers.
13 Potato Leaf Disease Detection and Classification Using VGG16 151
Fig. 1 Architecture
4 Modules
4.1 Image Processing
Image preprocessing is the practice of performing image processing on digital images
using computer algorithms. By using a specialized algorithm to analyze the image,
we can identify the plant. With a particular algorithm, we apply a similar strategy for
picture processing and detection. In this process, the image quality is crucial since
without a clear image, the algorithm would not work.
4.2 Data Splitting
Data splitting is a method of separating data into two or more parts. Testing or
evaluating the data in one part and training the model in the another portion of a two-
part split is common. It helps to guarantee the precision of procedures that use data
models, such as machine learning. A new approach can be used to effectively separate
data for the identification of plant leaf diseases. First, a dataset with pictures of both
152 V. L. Adluri et al.
Fig. 2 Transfer learning mechanism
Fig. 3 VGG 16 architecture
13 Potato Leaf Disease Detection and Classification Using VGG16 153
healthy and sick plant leaves is acquired. A wide and representative collection of
photos that includes various plant types, illnesses, and environmental circumstances
must be developed. Training, validation, and testing are the three main subsets that
can be extracted from the dataset. The training set makes up the majority of the data,
often between 70 and 80%. The model is trained using it to discover the fundamental
patterns of both healthy and sick leaves.
4.3 Feature Extraction
Feature extraction is a fundamental step in deep learning that involves identifying
and extracting relevant patterns or features from raw data. These extracted features
represent the most important aspects of the data and are used as input for subsequent
layers in a deep neural network. Through a sequence of convolutional and pooling
layers, CNNs are specially made to capture the spatial and hierarchical information
contained in images. The qualities of the leaf are fully represented by integrating
these various elements. With the use of this feature set, machine learning or pattern
recognition algorithms may accurately classify and identify leaf diseases. Effective
crop management and disease control techniques are aided by automated systems’
quick and accurate ability to recognize and diagnose plant diseases.
An overview of how feature extraction functions in CNNs is given below:
Convolutional Layers: Convolutional filters are used in the first layers of a CNN
to scan the input image in a sliding window fashion. Each filter creates a feature
map by performing element-wise multiplication on a tiny portion of the input image
using its weights. At various spatial scales, these filters record a variety of visual
patterns, including edges, textures, and forms. CNNs may simultaneously extract
several different kinds of information by using several filters.
Pooling Layers: By condensing and lowering the spatial dimensions of the feature
maps, pooling layers down sample the data. Max pooling, which selects the highest
value within a narrow window as the typical value for that area, is a common tech-
nique. By establishing translation invariance by pooling, the CNN becomes more
resistant to minute changes or variances in the input.
Fully Connected Layers: These are used in categorizing the learned characteristics
in the final stages of a CNN. These layers link every neuron from the prior layer to
the following layer, enabling the network to understand how the extracted features
relate to the intended job, such as classifying or identifying particular objects.
154 V. L. Adluri et al.
4.4 Data Classification
Data classification is essential for identifying plant leaf diseases. The next stage
is to categorize the leaf images into various disease groups after the preprocessing
methods have been applied. Each image of a leaf is classified according to the severity
of the disease present in that leaf.
For this objective, a variety of classification techniques, from basic machine
learning algorithms to deep learning models, can be used. These algorithms create
a predictive model using the preprocessed leaf images and the related disease diag-
noses. The data patterns and attributes that distinguish healthy leaves from diseased
ones and distinguish between various types of diseases are captured by the model.
In conventional machine learning methods, features are often extracted utilizing
methods from the preprocessed images. Convolutional neural networks (CNNs) are
an acclaimed type of deep learning used for various applications, including the detec-
tion and classification of plant diseases by analyzing leaf health. In a CNN, the
layers are grouped into different functional categories, with one or more convolu-
tional layers followed by subsampling layers and fully connected layers. Each feature
layer gets a small input feature set from the previous layer. CNNs have been utilized
with great success in identifying and classifying the various plant leaf diseases, which
is crucial for monitoring plant health in agriculture.
4.5 Testing
Testing is the procedure of assessing the efficiency and precision of a trained CNN
model on unobserved data. It is a critical stage in the creation and application of CNNs
since it enables us to evaluate how well the model generalizes to novel, unexplored
data and measures its efficiency in completing the task at hand, such as segmentation,
object recognition, or picture classification. In the testing phase, the trained CNN
model is applied to a different dataset known as the test set, which contains instances
not utilized in training or validation. This division is sure that the performance of the
model is estimated using data that it has never seen before, resulting in an objective
assessment. The confusion matrix shown in Fig. 4 helps us in finding the accuracy
of the proposed model with the following four parameters.
True Positives: Expected correct and output also correct.
True Negative: Expected wrong and output also wrong.
False Positive: Expected correct but the output was wrong.
False Negatives: Expected wrong but correct.
13 Potato Leaf Disease Detection and Classification Using VGG16 155
Fig. 4 Confusion matrix for a dataset
5 Results
In the proposed effort, experiments make use of the PlantVillage dataset. Work is
done on potato plant species. The technique can identify three different kinds of
diseases of a potato leaf: early blight, late blight, and healthy leaves. The VGG16
model has shown effectiveness in the diagnosis of plant leaf diseases. A DCNN
architecture called VGG16 is renowned for excelling at image categorization tasks.
The VGG16 model can learn detailed patterns and features linked to various types of
illnesses by training it on a large number of identified leaf photos. The VGG16 model
can properly categorize and identify the presence of disease with a high degree of
accuracy when given new leaf images. Early identification and prompt response are
made possible by this, which results in efficient disease management techniques and
increased crop health. Utilizing VGG16 to identify plant leaf diseases demonstrates
how deep learning models have the ability to revolutionize agricultural methods and
guarantee sustainable food production. The predictions of VGG 16 model are shown
in Figs. 5 and 6. Training and validation accuracy, training and validation loss of the
model are shown in Fig. 7.
156 V. L. Adluri et al.
Fig. 5 First image to predict
Fig. 6 Potato early blight
leaf
13 Potato Leaf Disease Detection and Classification Using VGG16 157
Fig. 7 Training and validation accuracy and training and validation loss
6 Conclusion
The suggested model uses convolutional neural networks to create a plant disease
detection system. As a classifier and feature extractor, VGG 16 is employed. On
different varieties of potato plants, experiments are done. By adding more data and
experimenting with various optimizers, deep learning-based architecture delivers
notable results that can be further enhanced.
References
1. Gaikwad VP, Musande V (2023) Potato plant leaf disease detection using CNN model. Eur
Chem Bull 12(1):516–527
2. Kothari D, Mishra H, Gharat M, Pandey V, Thakur R (2022) Potato leaf disease detection using
deep learning 11(11) November 2022
3. Bangal A, Pagar D, Patil H, Pande N (2022) Potato leaf disease detection and classification
using CNN 3(5):1510–1515, May 2022
4. Joshi K, Awale R, Ahmad S, Patil S, Pisal V Plant leaf disease detection using computer vision
techniques and machine learning. ICACC
5. Islam A, Sikder H (2022) A deep learning approach to classify the potato leaf disease 143–155
6. Moharekar TT, Urmila RP, Ombase R, Moharekar TJ Detection and classification of plant leaf
diseases using convolutional neural networks and streamlit
158 V. L. Adluri et al.
7. Chen J, Deng X, Wen Y, Chen W, Zeb A, Zhang D (2022) Weakly supervised learning method
for the recognition of potato leaf diseases. 21 December 2022
8. Rampurkar V, Kunika S, Todmal M, More A, Lakhotiya I (2022) An approach towards disease
detection in potato leaf. December 2022
9. Nishad MAR, Mitu MA, Jahan N (2022) Predicting and classifying potato leaf disease using
K-means segmentation techniques and deep learning networks. Proc Comput Sci 212:220–229.
ISSN 1877-0509. https://doi.org/10.1016/j.procs.2022.11.006
10. Kowshik B, Savitha V, Madhav N, Karpagam G, Sangeetha K (2021) Plant disease detection
using deep learning. March 2021
11. Singh A, Kaur H (2021) IOP Conf Ser: Mater Sci Eng 1022:012121
12. Kulkarni P, Karwande A, Kolhe T, Kamble S, Joshi A, Wyawahare M (2021) Plant disease
detection using image processing and machine learning, May 2021
13. Gajula Ramesh DD, William Albert D, Ramu G (2020) Detection of plant diseases by analyzing
the texture of leaf using ANN classifier
14. Sholihati RA et al (2020) Potato leaf disease classification using deep learning approach. In:
2020 international electronics symposium (IES), pp 392–397
15. Tiwari D, Ashish M, Gangwar N, Sharma A, Patel S, Bhardwaj S (2020) Potato leaf diseases
detection using deep learning. In: 2020 4th international conference on intelligent computing
and control systems (ICICCS). Madurai, India, pp 461–466. https://doi.org/10.1109/ICICCS
48265.2020.9121067
