Weakly-supervised learning method for the recognition of potato leaf diseases.

Area of Science:

• Agricultural Science
• Computer Vision
• Machine Learning

Background:

• Potatoes are a vital global food crop, vulnerable to various diseases that impact yield.
• Early disease detection is essential for preventing crop loss and ensuring food security.

Purpose of the Study:

• To develop a weakly-supervised learning approach for accurate potato plant disease identification.
• To enhance the MobileNet V2 architecture for improved detection of subtle disease lesions.

Main Methods:

• Utilized transfer learning to fine-tune the MobileNet V2 network.
• Integrated atrous convolution, SPP module, and a hybrid attention mechanism (channel and spatial) for feature extraction.
• Employed a weakly-supervised learning strategy for disease identification.

Main Results:

• Achieved an average recall rate of 91.99% on a public dataset for potato disease recognition.
• Demonstrated high performance in field scenarios with average accuracy of 97.33% and specificity of 98.39% on a local dataset.
• Outperformed existing methods in identifying potato plant diseases.

Conclusions:

• The proposed weakly-supervised learning approach is effective and feasible for potato plant disease identification.
• The enhanced MobileNet V2 architecture with attention mechanisms significantly improves disease detection capabilities.
• This method offers a promising solution for early disease diagnosis in potato cultivation.