Improving the maize crop row navigation line recognition method of YOLOX

Area of Science:

• Agricultural Engineering
• Computer Vision
• Machine Learning

Background:

• Accurate maize crop row identification is vital for autonomous agricultural machinery navigation.
• Lighting variations and complex field conditions pose significant challenges to existing detection methods.

Purpose of the Study:

• To develop an optimized YOLOX-Tiny model for robust maize crop row navigation line detection.
• To enhance the accuracy and efficiency of intelligent weeding machinery guidance systems.

Main Methods:

• An optimized YOLOX-Tiny model incorporating adaptive illumination adjustment and multi-scale prediction.
• Integration of visual attention mechanisms (Efficient Channel Attention, Cooperative Attention) and Fast Spatial Pyramid Pooling.
• Utilized Coordinate Intersection over Union loss function and least squares method for crop row fitting.

Main Results:

• Achieved 92.2% average precision with a detection time of 15.6 ms, a 16.4% improvement over the original model.
• Reduced model size by 7.1% to 18.6 MB.
• Demonstrated efficient crop row fitting with 42 ms processing time and 0.59° average angular error.

Conclusions:

• The enhanced YOLOX-Tiny model provides a highly accurate and efficient solution for maize crop row navigation.
• This technology significantly supports intelligent weeding machinery, promoting sustainable agriculture and increased crop yields.