Automatic grading evaluation of winter wheat lodging based on deep learning.

Area of Science:

• Agricultural Engineering
• Remote Sensing
• Computer Vision
• Machine Learning

Background:

• Winter wheat lodging significantly impacts crop yield and quality, necessitating accurate assessment for agricultural insurance and seed selection.
• Traditional methods for evaluating lodging are labor-intensive, subjective, and lack reliability, hindering efficient loss assessment.

Purpose of the Study:

• To develop and validate a novel deep learning model for accurate and automated grading of winter wheat lodging.
• To quantitatively assess lodging degree and area using unmanned aerial vehicle (UAV) remote sensing data.

Main Methods:

• A classification-semantic segmentation multitasking neural network, MLP_U-Net, was designed based on the U-Net architecture with improved MLP modules.
• The model employed a common encoder for robustness and utilized winter wheat lodging images captured by UAVs at varying altitudes.
• Datasets were created from images of 82 winter wheat varieties, enabling segmentation and classification of lodging severity and extent.

Main Results:

• MLP_U-Net demonstrated superior performance, achieving high accuracies in grading winter wheat lodging degree (96.1%) and lodging area (92.2%) at a UAV flight height of 30m.
• At a UAV flight height of 50m, the model still provided accurate grading, with accuracies of 84.1% for lodging degree and 84.7% for lodging area.
• The model proved robust and efficient, especially in small sample datasets, outperforming traditional methods.

Conclusions:

• The MLP_U-Net model offers a highly accurate, robust, and efficient solution for winter wheat lodging grading using UAV remote sensing.
• This technology provides valuable technical references for assessing winter wheat disaster severity and agricultural losses, improving insurance claim processing and seed selection.