Modified Resnet50 architecture for plant disease detection

Abstract

The early detection and diagnosis of plant diseases are essential for improving agricultural productivity and ensuring global food security. In recent years, deep learning techniques have demonstrated significant potential in enhancing the accuracy of plant disease identification. This study presents a modified ResNet50 architecture specifically designed for plant disease detection. The proposed model incorporates advanced features, including attention mechanisms, adaptive pooling layers, and feature recalibration techniques, to enhance its ability to identify diseased plant leaves from images. These modifications significantly improve the model’s capacity to recognize intricate patterns and subtle disease variations. Additionally, adaptive learning strategies utilizing large-scale datasets such as ImageNet have been employed to fine-tune the model for improved performance. The effectiveness of the modified ResNet50 architecture has been evaluated through extensive experimentation on multiple datasets, including PlantVillage and custom datasets. Comparative analysis with existing state-of-the-art approaches confirms that the proposed model achieves higher accuracy, robustness, and efficiency in detecting various plant diseases across different species and environmental conditions. Furthermore, this research examines the interpretability of model predictions and highlights potential directions for future advancements and real-world applications in smart agriculture.