In this paper,an improved YOLOv5 based submarine target detector is proposed to solve the problems that traditional submarine target detection lacks robustness to complex backgrounds and noises,is sensitive to changes in illumination and viewing angle,and is difficult to deal with large scale datasets.With the C3_Transformer structure,the global context modeling ability of the features and the long range dependency capturing ability are effectively improved.And the simOTA algorithm is employed to address the issue of imbalanced positive and negative samples in anchor based algorithms,thereby enhancing the model′s learning capabilities for small targets and challenging samples.Additionally,the decoupledhead approach is utilized to overcome the mutual exclusivity problem between classification and position prediction tasks,resulting in improved detection accuracy and robustness.The experimental results show that compared to the original YOLOv5,the improved model shows significant advancements in terms of Precision,Recall,mAP@0.5,and mAP@0.5∶0.95,with improvements of 2.8%,10.9%,3.8%,and 14.7% respectively,which indicates that the improved model achieves notable progress in terms of accuracy,recall rate,and average precision at different confidence thresholds in submarine target detection.Furthermore,the improved model effectively addresses the issues of "missed detection" and "false detection" in the actual detection task.