Plant Leaf Disease Detection using MobileNetV2

Abstract

Plant diseases significantly hinder agricultural productivity worldwide, making early detection and accurate diagnosis essential to safeguard crop yields and food security. An efficient and lightweight deep learning method designed for classifying plant diseases using the PlantVillage dataset is presented. MobileNetV2, a pre-trained convolutional neural network optimized for efficient operation on mobile and embedded platforms, is employed. The dataset comprises 54,305 images across 38 classes, The model applies preprocessing steps like resizing, normalization, and data augmentation to improve model generalization. The training process used the Adam optimization algorithm in combination with the categorical cross-entropy loss function, achieving a validation accuracy of 94.32%. The evaluation process considers metrics such as precision, recall, and the F1-score to confirm strong performance in classification tasks. The proposed system’s compact design, high accuracy, and low computational demands make it suitable for mobile or web-based agricultural tools, enabling farmers to obtain rapid and reliable diagnoses in real time, thereby supporting better decision-making and improved crop health management.