Deep Learning-Driven Sustainable Strategies for Potato Leaf Disease Detection to Enhance Crop Yield and Soil Fertility

Abstract

Potato plants are very important in world food security but also have a high susceptibility to diseases like early blight and late blight, which increase crop loss and hamper stable production practices. In this study, an early potato leaf disease detection method using deep learning is proposed, apply transfer-learning methods to improve model accuracy and efficiency. The network was trained with Potato Disease Leaf Dataset from Kaggle, with leaves classified as healthy, early blight, and late blight. Three pre-trained Convolutional neural networks VGG16, GoogLeNet, and ResNet-50 were employed and compared. ResNet-50 performed best among them, with superior 100% training accuracy, 99.75% validation accuracy, and 99.26% testing accuracy, then GoogLeNet and VGG16. All the models were excellent on Precision, Recall, F1-score, and Misclassification Rate. The system enables farmers to make timely and precise interventions, reducing their use of agrochemicals, improving the resistance of their crops, and contributing to sustainable agriculture practices in line with the United Nations' Sustainable Development Goals. By combining modern Artificial Intelligence (AI) technologies with green approaches, this work emphasizes the potential to increase crop production, save soil fertility, and achieve long-term agricultural productivity.