Ginseng Seed Quality Detection Based on YOLO-GS

Seed quality is a crucial factor in determining yield before sowing. Ginseng seeds undergo several processes before sowing, including picking, washing, and germination. The germination process is susceptible to damage or failure, which can directly impact the final yield of subsequent cultivation. Therefore, precise and reliable quality inspection and screening must be done before sowing to ensure a high germination rate. Based on YOLOv11n, this study proposes the YOLO-GS model to test the quality of ginseng seeds. Firstly, a SELP module was designed to enhance the network's ability to focus on the key features of ginseng seeds and improve the model's detection accuracy. Secondly, the Channel Prior Convolution Attention (CPCA) mechanism was introduced into the backbone network to dynamically assign attention weights to the feature map in both the channel and spatial dimensions, thereby enhancing the network's ability to extract features from the target. Thirdly, the C3k2 structure in the backbone was improved to account for both local feature extraction and global dependency modeling, thereby enhancing the model's accuracy. Finally, a Convolutional Attention Module (CloFormerAttnConv) based on the multi-frequency position-sensitive attention mechanism in C2PSA was introduced to achieve a dual perception of local details and global semantics while maintaining computational efficiency and improving feature extraction capabilities. The experimental findings demonstrated that the YOLO-GS model attained 97.7% mAP@0.5, with Precision, Recall, F1-Score, and mAP@0.5:0.95 reaching 96.7%, 96.4%, 90.5% and 90.3%, respectively. The model has only 4.2 million parameters. When deployed on the Jetson edge device, the model inference time is 0.6ms, providing an effective solution for real-time target detection tasks in the application of seed quality assessment of ginseng. In conclusion, the YOLO-GS model will be applicable for the precise detection of ginseng seed quality.