Enhanced YOLOv11n: A Method for Potato Peel Damage Detection

Area of Science:

• Agricultural Engineering
• Computer Vision
• Food Science

Background:

• Potato skin damage during harvest and transport compromises storage, quality, and safety.
• Rapid detection of potato damage is essential for automated sorting processes.

Purpose of the Study:

• To develop an enhanced YOLOv11n algorithm for accurate and efficient detection of potato peel damage.
• To improve the performance of automated potato sorting systems by identifying damaged produce.

Main Methods:

• Integration of MLCA attention mechanism and Re-Calibration FPN for enhanced feature extraction and multi-scale fusion.
• Implementation of EIEStem module and SobelConv for improved feature representation of potato skin damage.
• Optimization of the detection head with AFPN and introduction of WloU loss function for precise boundary regression.

Main Results:

• The enhanced YOLOv11n algorithm achieved significant improvements: 6.9% in mAP@0.5, 15% in mAP@0.5:0.95, 7.4% in precision (P), and 11.8% in recall (R) compared to the benchmark.
• Demonstrated superior performance over existing mainstream algorithms in comprehensive evaluations.

Conclusions:

• The proposed enhanced YOLOv11n algorithm effectively detects potato peel damage.
• The advancements in feature extraction, fusion, and loss function contribute to superior detection accuracy and reduced errors.