Enhanced <scp>YOLOv11n</scp> : A Method for Potato Peel Damage Detection

Potatoes are susceptible to skin damage during harvest and transportation, which impacts storage capabilities, increases the risk of microbial contamination and decay, and diminishes both quality and food safety. Rapid detection of potato peel damage is crucial for the subsequent automatic sorting process. To address this, this paper introduces an enhanced YOLOv11n target detection algorithm for identifying damaged potatoes. The primary enhancements include: integrating the MLCA attention mechanism into the C3k2 module of the Backbone network to combine channel and spatial attention, thereby enhancing the network's ability to detect skin damage features; proposing a Re-Calibration FPN to refine the Neck component, optimizing multi-scale feature fusion through bidirectional fusion and an adaptive attention mechanism to accurately define the area of skin damage and improve the effectiveness of multi-scale feature fusion; replacing the first two convolution layers of the Backbone with the EIEStem module, and combining SobelConv with convolutional branches to strengthen the feature representation of potato peel damage images; optimizing the detection head with AFPN to narrow the semantic gap, enhance feature fusion, and further improve the detection of potato peel damage. The WloU loss function is introduced to enhance the precision of boundary regression and reduce the occurrence of missed and false detections of skin damage. Comparative and ablation experiments demonstrate that the improved algorithm outperforms the benchmark YOLOv11n model by 6.9%, 15%, 7.4%, and 11.8% in terms of [email protected], [email protected]:0.95, precision rate (P), and recall rate (R), respectively. Compared to existing mainstream algorithms, the proposed algorithm exhibits superior performance in comprehensive comparisons, fully validating its effectiveness in the task of potato peel damage detection.