【摘要】 为提升船舶在通航场景下对海上目标的识别效果，保障其航行安全性，针对传统目标检测算法存在的漏检和误检问题，提出一种改进的海上目标检测算法。基于主流的目标检测算法YOLOv5，通过引入有效挤压激励（EffectiveSE）注意力机制、多尺度上下文增强模块和多梯度路径卷积结构等3种手段对该算法进行改进。采用通航船舶采集的海上目标图像数据集进行消融试验，结果表明，3种改进手段均能有效提高模型的识别精确率和召回率，明显改善原算法存在的漏检和误检问题。对该改进算法与其他常用的海上目标检测算法进行对比试验，结果显示，改进后的算法对海上目标的平均识别准确率可提升至89.6%，单帧图片检测时间为8.6ms，相比其他算法检测效果明显提升。改进算法提高微小目标识别精度，实现对通航环境下海上目标的实时高精度检测，满足实时目标检测需求。更多还原

【Abstract】 The maritime navigation environment is complex, and most of the safety accidents that occur in ships during navigation are due to the lack of effective perception of the environment. Based on the mainstream target detection algorithm YOLOv5, an improved algorithm by introducing the effective squeeze and excitation(EffectiveSE) attention mechanism is proposed, multi-scale context enhancement module and multi-gradient path convolution structure. The results of ablation experiments using maritime target image datasets collected by navigable vessels show that all three improved means can effectively improve the accuracy and recall rate of the model, and significantly improve the problem of missed and false detection existing in the original algorithm. Comparison tests between the improved algorithm and other commonly used maritime target detection algorithms show that the average recognition accuracy of the improved algorithm for maritime targets can be increased to 89.6%, and the detection time of a single frame is 8.6 ms,which is a significant increase in the detection effect compared to other algorithms. The improved algorithm improves the recognition accuracy of tiny targets, realizes the real-time high-precision detection of marine targets under navigable environment, which meets the requirement of real-time target detection.更多还原