Crop Disease Identification and Severity Grading Model Based on ResNet50 and Multi-Task Learning

Abstract

This paper proposes a novel approach for crop disease identification and severity grading based on deep learning models. The focus lies on utilizing lightweight convolutional neural networks (e.g., ResNet50) as backbone models for disease classification, while enhancing recognition performance under limited data through transfer learning and strong data augmentation techniques. First, a customized dataset was constructed and processed through data cleaning. A deep learning framework was employed to fine-tune pre-trained models, accomplishing the crop disease classification task. Second, a multi-task model was developed to simultaneously handle disease classification and severity grading tasks. This model efficiently captures complex visual patterns by sharing the feature extraction backbone network. Finally, strong data augmentation techniques were introduced to enhance the model's robustness under small-sample conditions. The model's primary advantages include improved generalization capability, maintained high accuracy, and effective application across different crops, disease types, and severity levels. This research provides a more accurate and interpretable model for crop disease management.