Page 53 - Kaleidoscope Academic Conference Proceedings 2024

POTATO PLANT LEAF DISEASE DETECTION USING CUSTOM CNN DEEP NET: A STEP
TOWARDS SUSTAINABLE AGRICULTURE

2
3
Kyamelia, Roy ; Subharthy, Ray ; Tapan Kumar, Pal ; Sheli, Sinha Chaudhuri 2
1
1 Dept. of Electronics and Tele-Communication Engineering, Siliguri Govt. Polytechnic, Siliguri, West Bengal, India.
Email: kyamelia2015@gmail.com
2 Dept. of Electronics and Tele-Communication Engineering, Jadavpur University, Kolkata, West Bengal, India.
Email: subharthiray126@gmail.com, shelisinhachaudhuri@gmail.com
3 Kanyapur Polytechnic, Asansol, West Bengal, India. Email: pal.tapan.k@gmail.com

ABSTRACT
1. INTRODUCTION
Detecting potato plant leaf diseases using convolutional
neural networks (CNNs) is a significant step towards Conference Automation with computer vision enabled
sustainable agriculture. Computer vision-based automated with deep learning in agriculture revolutionizes
disease detection in agriculture is essential for the early efficiency and productivity from planting to harvesting,
detection and treatment of plant diseases, assisting farmers ensuring sustainable food production for the future.
in minimizing crop losses, maximizing yields, and Convolutional neural networks (CNNs) to detect leaf
guaranteeing food security. This study proposes a novel diseases in potato plants is a big step towards sustainable
method for detecting potato leaf disease using a customized agriculture. Farmers can reduce the need for widespread
5-layer Convolutional Neural Network (CNN). The model's pesticide application by implementing targeted
performance is compared with MobileNet and a 4-layer treatments in response to early disease identification. By
CNN as the backbone architectures. Based on experimental doing this, chemical use is reduced while simultaneously
results, the 5-layer customized CNN achieves an accuracy protecting soil quality and avoiding environmental
rate of 97.16%, which is significantly higher than that of contamination. Furthermore, early disease detection
the 4-layer CNN (73.21%) and MobileNet (78.43%). promotes sustainable farming methods, ensures food
Additionally, the 5-layer CNN model shows promising security, and prevents yield loss. All things considered,
results for other evaluation metrics, including F1 score using technology to detect diseases helps make potato
(97.18%), recall (97.16%), and precision (97.24%). farming more ecologically friendly and sustainable.
Furthermore, out of all the models that were tested, the 5- Potato (Solanum tuberosum) is one of the most important
layer CNN model shows the least amount of loss. Using a food crops worldwide, playing a vital role in global
threshold of 0.6, and the custom CNN as backbone agriculture and food security. Not only is it a staple food
architecture to Faster R-CNN (FRCNN) the model achieved for millions of people, but it also serves as a significant
an Intersection over Union (IoU) of 0.76 for disease source of income for farmers and contributes
detection. Additionally, a comparative study of various significantly to the economies of many countries.
optimizers (Adam, SGD, Adadelta, and AdamW) and loss However, potato plants are susceptible to various
functions is done; the Adam optimizer and the unique 5- diseases, which can have detrimental effects on both
layer CNN model yielded the best results. This study yield and quality. Early detection and management of
advances automated methods for detecting potato leaf these diseases are crucial for sustainable potato
diseases, offering a dependable and effective way to identify production.
diseases early in agricultural settings.
In recent years, there has been a growing interest in
Keywords – Computer Vision, Convolution Neural using technology, particularly deep learning
Network (CNN), Sustainable agriculture, Early Blight, techniques, for the early detection and management of
Late Blight plant diseases. In this paper, we propose a novel
approach for potato plant leaf disease detection using a
custom-designed Convolutional Neural Network

978-92-61-39091-4/CFP2268P @ITU 2024 – 9 – Kaleidoscope