基于三种组合改进YOLOv8模型的马铃薯芽眼识别方法OA

To achieve rapid and accurate identification of potato seed tuber eyes,a potato eye recognition method based on three combined improved YOLOv8 models was proposed.Firstly,the Bi-directional Feature Pyramid Network(BiFPN)was replaced,and the P2-level small target detection layer was integrated to enhance the model's ability to detect multi-scale eye targets and improve the model's robustness.Secondly,the Convolutional Block Attention Module(CBAM)was introduced to improve the model's detection performance for small target eyes.Finally,the Non-Maximum Suppression(NMS)algorithm was optimized to increase the recall rate.The results showed that the model achieved a precision(P)of 97.34％,a recall(R)of 94.23％,a harmonic mean F1-Score of 95.76％,a mean Average Precision(mAP@0.5:0.95)of 95.47％,and the recognition time for a single image was 16.3 ms on average.This model performed excellently in the potato eye recognition task with stronger anti-interference capability.It could provide an important theoretical reference for eye recognition in the intelligent cutting process of potato seed tubers and hold significant practical value for promoting the full-process mechanization development of the potato industry.