| 摘要 |
[Objectives] This study was conducted to propose an improved YOLOv8s-based potato identification algorithm, in order to address issues such as low detection precision and high miss detection rate in potato identification tasks. [Methods] The backbone network was replaced with InceptionNeXt, and the CBAM dual attention mechanism was introduced, to enhance the model's multi-scale feature extraction capability. [Results] The improved YOLOv8s algorithm achieved an identification precision of 94.55%, a recall of 85.34%, and an F1-score of 87.37% in potato identification. Compared with the original algorithm, it improved precision by 7.40%, recall by 2.71%, and F1-score by 2.56%. The average processing time per image was reduced by 0.12 s compared with the unimproved algorithm. The results of simulation tests showed a success rate of 98.20% in 2 000 simulated identifcation tests. [Conclusions] This study provides a high-precision and robust solution for potato identification tasks. |
