Potato Leaves Blight Disease Recognition and Categorization Using Deep Learning

Abstract

Potato cultivation is vital in numerous countries, contributing to food security and economic value. However, crop diseases, particularly early and late blight, pose significant challenges to potato production. The accurate diagnosis of these diseases remains unclear to many individuals. This study leverages the increasing penetration of smartphones and recent advancements in deep learning to develop a Convolutional Neural Network (CNN) model for real-time detection of early and late blight in potatoes. The dataset was pre-processed by normalizing, dividing, and extracting images using the Python data processing library. The approach incorporates slight variations in the network layers to optimize the model's performance. The method was evaluated using classification optimizers, metrics, and loss functions and further refined using layer-by-layer TensorBoard analysis. Hyperparameters such as features, labels, validation split, batch size, and training epochs were carefully selected. The final model demonstrated promising results, achieving an accuracy of 96.09% on the survey dataset. Experimental findings highlight the approach's potential for automatically detecting both early, late blight and healthy, thereby significantly improving the accuracy of disease diagnosis.