Research on cabbage transplanting status detection and operation quality evaluation in complex environments based on...

Area of Science:

• Agricultural Engineering
• Computer Vision
• Machine Learning

Background:

• Crop transplanting quality significantly impacts plant survival and yield.
• Mechanized transplanting faces challenges with manual inspection efficiency and traditional algorithm accuracy.
• Open-field environments present difficulties for current seedling detection and tracking methods.

Purpose of the Study:

• To develop a robust detection-and-tracking method for assessing cabbage transplanting quality in real-world conditions.
• To enhance the accuracy and stability of identifying various transplanting states (normal, soil-buried, bare-root).
• To achieve precise and reliable counting of transplanted seedlings for improved agricultural management.

Main Methods:

• An improved YOLOv10-TQ network with triplet attention and QFocal Loss-cross-entropy was utilized for enhanced seedling state detection.
• A lightweight MobileViT feature extractor was integrated into the DeepSort algorithm for improved target tracking.
• A line-crossing counting strategy was implemented for accurate seedling enumeration and identity de-duplication.

Main Results:

• The proposed method achieved a mean average precision (mAP) of 86.3% for seedling state detection.
• An average counting accuracy of 97.8% was recorded on a custom cabbage transplanting dataset.
• The system demonstrated superior detection accuracy, tracking stability, and counting precision over traditional methods.

Conclusions:

• The developed AI-driven approach offers a significant advancement in the intelligent quality evaluation of cabbage transplanting.
• The method provides a strong technical foundation for data-driven decision-making in agricultural machinery systems.
• A visualization system was created to monitor transplanting status, promoting precision agriculture.