Fine-Grained Image Classification for Crop Disease Based on Attention Mechanism

Area of Science:

• Agricultural Science
• Computer Vision
• Machine Learning

Background:

• Fine-grained image classification is difficult due to subtle feature identification.
• Crop disease classification faces challenges from visual disturbances like light, fog, and overlap.
• Accurate classification requires models to focus on discriminative regions in complex scenes.

Purpose of the Study:

• To explore the influence of crop leaf image features on classification results.
• To develop a classification model that focuses on discriminative regions and improves accuracy in complex scenes.
• To propose a novel attention mechanism for effective utilization of informative image regions in crop disease classification.

Main Methods:

• Utilized transfer learning to construct fine-grained image classification models for crop disease.
• Developed and integrated a novel attention mechanism into the NASNetLarge model.
• Trained and evaluated models on a dataset of 58,200 crop leaf images (14 crops, 37 categories).

Main Results:

• The NASNetLarge model with the proposed attention mechanism achieved the best performance.
• Achieved a high F1 score of 93.05% in fine-grained classification of crop disease images.
• Demonstrated the effectiveness of the attention mechanism in improving classification accuracy.

Conclusions:

• The proposed attention mechanism significantly enhances fine-grained classification of crop disease images.
• The approach is effective in handling visual disturbances and subtle feature variations.
• This method provides a robust solution for automated crop disease identification.