In response to the insufficient feature extraction caused by factors such as the complex underwater en-vironment,low visibility,and uneven light distribution,a YOLOv8n-DSP underwater biological target detection method based on the improvement of YOLOv8n is proposed.Firstly,a DBB(Diverse Branch Block)module is added to the feature extraction backbone network of YOLOv8n and fused with the C2f module to combine multiple branches of different scales and complexities,enriching the feature space and strengthening the network's feature extraction capability.Then,a weighted attention mechanism(SimAM)is added before spatial pyramid pooling to adaptively adjust the model's focus and enhance its representation ability in complex scenes.Finally,to further improve the localization accuracy of the object detection boxes,the existing CIoU is replaced with PIoU(Power-ful-IoU),resulting in better regression of the object detection boxes.Experimental results show that the YOLOv8n-DSP network improves mAP@0.5 and mAP@0.5:0.95 by 2%and 2.2%,respectively,compared to the original YOLOv8n network.
维普
万方数据