High-Resolution Leaf Disease Classification Using Deep Learning Advancing Agricultural Technology Through Innovative Approaches

Abstract

Accurate classification of plant diseases is essential for effective agricultural management and crop protection. Leaf diseases pose serious threats to crop growth, leading to significant challenges and financial losses for farmers. This study presents an approach leveraging combined deep learning techniques for the classification of tomato leaf diseases using transfer learning. We evaluated the performance of two widely used convolutional neural network architectures: DenseNet121 and MobileNetV2, exploring the impact of freezing and unfreezing layers during training, resulting in four distinct model configurations. By selectively unfreezing layers, we enhanced fine-tuning, leading to improved model performance. Our experiments revealed that DenseNet121 with unfrozen layers achieved the highest classification accuracy, with a validation accuracy of 99.88% and a test accuracy of 99.45%. These results underscore that the proposed approach not only excels in accuracy but also provides a reliable and precise solution for managing and controlling leaf plant diseases, making it particularly effective for practical agricultural applications. © 2025 Elsevier B.V., All rights reserved.