摘要： 在渔港高点监控渔船目标的场景下，对渔船检测经常丢失和检测错误等问题，提出了一种基于改进Yolov5的渔船目标检测模型。首先通过Kmeans++算法对锚框重新聚类，选择适合渔船数据集的锚框尺寸；然后在Yolov5的骨干网络中融入CBAM注意力机制获取更多细节特征；再采用加权双向特征金字塔网络（BiFPN）代替原先的特征金字塔网络（FPN）+像素聚合网络（PAN）结构，快速进行多尺度特征融合；最后在检测尺度上去掉大目标的检测尺度，增加更小目标的检测尺度，改用新的三个检测尺度，提高了模型对小目标渔船的检测精度。结果显示：对比原Yolov5算法，改进后的算法精确度、召回率和平均精度均值均有所提升，分别提升29.5%、0.5%和4.5%，每秒检测帧数达到90.6，对渔船目标检测效果有大幅度改善。研究表明，改进后的Yolov5算法满足休渔期管控期间对渔船目标检测的准确性和实时性要求。

关键词: 渔船检测, Yolov5算法, CBAM注意力机制, 加权双向特征金字塔

Abstract: In the scene of monitoring fishing boat targets at the high point of the fishing port, a fishing vessel target detection model based on improved yolov5 is proposed to solve the problems of frequent loss and detection error of fishing vessel detection. Firstly, the anchor frame is re-clustered by the Kmeans++ algorithm, and the anchor frame size suitable for the fishing vessel data set is selected; Then, the weighted bidirectional feature pyramid network (BiFPN) is used to replace the original Feature Pyramid Networks (FPN) + Pixel Aggregation Network (PAN) structure for fast multi-scale feature fusion; Finally, the detection scale of large targets is removed, the detection scale of smaller targets is added, and three new detection scales are used to improve the detection accuracy of the model for small target fishing vessels. The results show that compared with the original yolov5 algorithm, the accuracy, recall and average accuracy of the improved algorithm are improved by 29.5%, 0.5% and 4.5% respectively, and the number of detection frames per second reaches 90.6, which greatly improves the effect of fishing boat target detection.The research shows that the improved yolov5 algorithm meets the accuracy and real-time requirements of fishing vessel target detection during the fishing moratorium.

Key words: fishing vessel detection, Yolov5 algorithm, CBAM attention mechanism, BiFPN