A small floating target detection algorithm on the water surface based on the improved YOLOv5s

In order to solve the problems of false detection and missed detection in the recognition of floating bottles on the water surface from the perspective of unmanned boats, an improved YOLOv5s floating small target detection algorithm was proposed based on the YOLOv5s algorithm. Firstly, the data augmentation was carried out for the original dataset Flow-Img, and the original dataset was expanded to avoid the phenomenon of overfitting of the model. Secondly, in order to improve the detection accuracy of the deep learning model for very small targets, a very small target detection layer is added on the basis of the three detection layers of YOLOv5s, and the detection head for large targets is removed to avoid the problem of prior frame allocation caused by data imbalance. Thirdly, the CBAM attention module was added to the backbone network to solve the problem of insufficient ability of the model to capture target feature information in the floating bottle detection task on the water surface. Finally, the Normalized Wasserstein Distance (NWD) regression loss function is introduced, and the IoU loss function and the NWD loss function are weighted and combined to form a comprehensive regression loss function, so as to further improve the accuracy and precision of the recognition of floating bottles on the water surface. Experimental results show that the mAP@0.5 value of the algorithm reaches 95.7% in the detection of floating bottles on the water surface, and compared with the original YOLOv5s algorithm, the mAP@0.5 of the improved YOLOv5s algorithm is increased by 2.6%, the mAP@0.95 is increased by 4.5%, and the number of model parameters is reduced by 61.9%. While achieving lightweight, the detection results of floating bottles on the water surface are more accurate, which provides an important technical reference for the detection of small floating objects on the water surface.