Comparative Analysis of Deep Learning Models for Potato Leaf Disease Detection

In India, agriculture plays a crucial role in ensuring food security and employment for its large population. The agricultural sector contributes significantly to the country's economy. Potato cultivation is prominent in India, with potato being a staple food, rich in essential nutrients, and having various indigenous and improved varieties. However, potato crops face challenges from fungal diseases like Early Blight and Late Blight, which can cause substantial economic losses. Currently, manual observation is the prevalent method for disease detection, but it has limitations. It is often impractical due to time constraints and limited expertise, leading to potential errors and inadequate treatment. This approach can result in crop failure due to delayed or incorrect interventions. Automatic plant leaf disease detection system utilizing technologies like Machine Learning, Deep Learning, and Computer Vision are being developed to promote timely and accurate detection. Among the varieties of techniques available, Deep Learning algorithms, and specifically Convolutional Neural Networks (CNNs), have emerged as formidable tools for disease classification. By autonomously learning complex features and patterns from images, CNNs demonstrate remarkable proficiency in distinguishing between healthy and diseased plant leaves, making them a pivotal component of modern agricultural innovation. This study focuses on using various self-built and pre-trained DL models for disease detection in potato leaves using two different verified datasets, and undertaking a comparative assessment of the different models. The research showed that out of the eight models used, ResNet50 gives the best accuracy for this experiment, followed by VGG16, CNN-KNN Hybrid, VGG19, SBCNN, InceptionV3, AlexNet, and CNN-SVM Hybrid.