Research and testing of a robot vision-based perception method for assessing corn sowing quality

Area of Science:

• Agricultural Engineering
• Computer Vision
• Precision Agriculture

Background:

• Manual inspection of corn sowing quality is inefficient, labor-intensive, and time-consuming.
• Automated methods are needed to improve the accuracy and speed of sowing quality assessment.

Purpose of the Study:

• To develop an automatic plant-spacing measurement method for corn seedlings using 3D machine vision.
• To establish a framework for automated sowing quality evaluation.

Main Methods:

• A mobile platform with a stereo camera acquired real-time RGB and depth data.
• The YOLOv11-Pose model detected plant keypoints for 3D reconstruction and distance calculation.
• A sowing-quality evaluation framework analyzed Quality of Feed Index (QFI), Multiple Index (MUL), Miss Index (MI), and coefficient of variation.

Main Results:

• The system achieved high accuracy in keypoint detection (mAP@0.5: 0.990, mAP@0.5:0.95: 0.989).
• Qualified indices for sowing quality were 77.83%, 80.36%, and 82.46% for different spacings.
• The system's sowing quality metrics aligned with manual measurements.

Conclusions:

• The proposed 3D machine vision method enables efficient, non-destructive detection of corn seedling spacing and sowing quality.
• This technology provides reliable support for precision sowing and field management.