【摘要】 针对当前海上环境复杂、噪声干扰严重及船舶检测存在漏检和误检等问题，提出一种基于注意力机制的TCS-YOLO船舶检测系统。该算法使用K-Means++聚类算法确定目标样本的锚框以提高先验框与船舶目标的尺寸匹配度；在YOLOv7的Neck部分引入Transformer Block以捕获全局信息和丰富的上下文信息；在YOLOv7的Head部分添加CA注意力机制，有助于模型更准确地定位和识别感兴趣的对象。并在自制的船舶数据集上进行试验，结果表明，该算法的平均精度均值达到70.5%，相比原始的YOLOv7算法值提高了5.1%，能更准确地检测船舶，满足在复杂海上环境中进行船舶检测的需求。更多还原

【Abstract】 In response to the issues of missed detection and false detection caused by the complexity of marine environments and severe noise interference in ship detection, a transformer coordinate-attention ship detection-YOLO(TCS) algorithm based on the attention mechanism is proposed. The TCS algorithm improves the matching degree between prior boxes and ship targets by using the K-Means++ clustering algorithm to determine the anchor boxes of target samples. Additionally, a transformer block in the neck part of YOLOv7 is introduced to capture global information and rich contextual information. In the head part of YOLOv7, the CA attention mechanism is added, which help the model accurately locate and recognize objects of interest. The experiments are conducted on a self-made ship dataset, and the mean average precision reaches 70.5%. The results demonstrate that the mAP of our algorithm is 5.1% higher than that of the original YOLOv7 algorithm, and it can detect ships more accurately, which can meet the needs of ship detection in complex marine environments.更多还原