Accurate Wheat Lodging Extraction from Multi-Channel UAV Images Using a Lightweight Network Model.

Area of Science:

• Agricultural Engineering
• Remote Sensing
• Computer Vision

Background:

• Wheat lodging significantly impacts agricultural management and disaster assessment.
• Current methods for lodging recognition suffer from low accuracy and poor real-time performance in complex field environments.

Purpose of the Study:

• To develop a highly accurate and real-time wheat lodging extraction model.
• To improve post-disaster agricultural management and assessment capabilities.

Main Methods:

• Constructed four-channel fusion images (RGB+DSM, RGB+ExG) from UAV-acquired RGB images.
• Proposed a Mobile U-Net model integrating lightweight networks, depthwise separable convolution, and U-Net architecture.
• Trained and evaluated the model using RGB, RGB+DSM, and RGB+ExG datasets.

Main Results:

• The RGB+DSM dataset achieved 88.99% accuracy, outperforming RGB and RGB+ExG.
• The Mobile U-Net model demonstrated superior performance over FCN and U-Net in parameters, speed, and accuracy.
• Achieved 27.3% and 33.3% faster processing speeds compared to FCN and U-Net, respectively.

Conclusions:

• The Mobile U-Net model utilizing RGB+DSM fusion offers a robust and accurate solution for wheat lodging extraction.
• This approach enhances agricultural disaster assessment and management efficiency.