A detection method for drowning personnel based on improved YOLOv7
Once a drowning accident occurs,timely detection of the drowning object and taking active corresponding measures play a crucial role in reducing the consequences of casualties.Therefore,automatic monitoring of drowning personnel has become particularly important.In the process of simulating the detection of personnel falling into the water in the wild,it was found that the natural water scene is complex,with problems such as small detection targets and large environmental interference.The existing benchmark model YOLOv7 has a high target miss rate.A ResNet ACmix has added to the backbone component of the YOLOv7 model,and an ACmix has added in the Head section to preserve the features collected by Backbone.It effectively extracts the fea-ture information of small targets of underwater personnel in complex scenes,enhances the model's feature perception and position information of small targets of underwater personnel,reduces the interference of complex water environments on feature extraction,and improves the pooling part in SPPCSPC.It ensures the positioning of small targets for drowning personnel in large and complex water areas,further reducing environmental interference and improving detection speed.Various experiments conducted on the con-structed dataset of drowning personnel have shown that compared with the YOLOv7 algorithm of the baseline network,the im-proved YOLOv7 algorithm has an average accuracy of 80.7%,an increase of 7%.The ablation experiment shows that the designed module can improve detection accuracy and visually display the detection results in different scenarios.The experiment verified the applicability of improved YOLOv7 in target detection of underwater personnel.
万方数据
维普
