YOLOV5 Underwater Object Detection Algorithm Based on Multi-aggregation and Multi-scale
Aiming at the higher requirements for the real-time and accuracy of detection of object such as holothurian,echi-nus and scallop in ocean ranch,a YOLOV5 underwater object detection algorithm based on multi-aggregation and multi-scale is proposed.The algorithm introduces the CIoU Loss loss function and DIoU-NMS non-maximum suppression method to improve the problem of missed detection of object in data-intensive scenarios.In order to improve the detection accuracy of small objects,a small object detection layer is added to the network.Then,the coordinate attention mechanism CoordAtt is introduced to extract the coordinate information of different levels of feature maps.Finally,the structure of the feature fusion network is improved,and the ex-tracted coordinate information is fused with the feature maps of the corresponding scale in the feature fusion network.In the experi-mental analysis,the indicators commonly used in object detection are introduced as the evaluation criteria,including the average ac-curacy(AP),average accuracy(mAP)and detection speed.The performance of the algorithm is verified through ablation experi-ments and compared with the experimental results of other latest one-stage algorithms.The experimental results show that the algo-rithm proposed in this paper at an advanced level.
万方数据
维普
