Early Plant Disease Classification System Based on Deep Neural Network Techniques

Abstract

Plant diseases pose a significant challenge to agricultural productivity because they can reduce crop yield and quality. Accurate and automated disease identification can support early diagnosis and improve crop management. This study presents a deep learning-based approach for classifying plant diseases from leaf images using the PlantVillage dataset. The dataset contains plant leaf images representing healthy and diseased conditions, which were preprocessed by resizing the images to 224 × 224 pixels, normalizing pixel values, and applying data augmentation to improve model robustness. Two deep learning approaches were evaluated: a custom Convolutional Neural Network (CNN) and DenseNet121 using transfer learning with ImageNet-pretrained weights. The models were trained using 70% of the dataset, while 10% and 20% were used for validation and testing, respectively. Experimental results show that the custom CNN achieved a validation accuracy of 98%, with a training accuracy of 97.36%, while DenseNet121 achieved a validation accuracy of 96.85% and a training accuracy of 98.23%. The results demonstrate that the custom CNN achieved higher validation accuracy than the DenseNet121-based transfer learning approach for the evaluated plant disease classification task. These findings indicate that a relatively compact CNN architecture can provide competitive performance for automated plant disease classification