Deep Learning for Plant Disease Detection

Abstract

Agriculture, an essential bedrock of human survival, continually grapples with the menace of plant diseases, culminating in substantial yield reductions. While conventional detection techniques remain widespread, they often entail laborious efforts and are susceptible to inaccuracies, underscoring the pressing need for more efficient, scalable, and immediate solutions. Our research explores the transformative capabilities of Deep Learning (DL) models, primarily focusing on Convolutional Neural Networks (CNNs) and MobileNet architectures in the early and precise identification of plant ailments. We augmented our exploration by incorporating eXplainable Artificial Intelligence (XAI) through GradCAM, which elucidated the decision-making process of these models, providing a visual interpretation of disease indicators in plant images. Through rigorous testing, our CNN model yielded an accuracy of 89%, a precision and recall of 96%, and an F1-score of 96%. Conversely, the MobileNet design showcased an accuracy of 96% but recorded slightly lesser precision, recall, and F1-scores of 90%, 89%, and 89%, respectively. Such results amplify the transformative role of DL in redefining plant disease detection methodologies, presenting a formidable counterpart to conventional techniques and ushering in an era of heightened agricultural security.