Gas Plume Target Detection in Multibeam Water Column Image Using Deep Residual Aggregation Structure and Attention Mechanism

A multibeam water column image (WCI) can provide detailed seabed information and is an important means of underwater target detection. However, gas plume targets in have no obvious contour are susceptible to the influence environments, equipment noises, other factors, resulting varied shapes sizes. Compared with traditional detection methods, this paper proposes improved YOLOv7 (You Only Look Once vision 7) network structure for detecting a WCI. Firstly, Fused-MBConv used replace all convolutional blocks ELAN (Efficient Layer Aggregation Networks) module form ELAN-F (ELAN based on block) module, which accelerates model convergence. Additionally, MBConv 3 × ELAN-M reduces number parameters. Both modules deep residual aggregation structures fuse multilevel features enhance expression. Furthermore, ELAN-F1M3 one block three blocks) backbone designed fully leverage efficiency modules. Finally, SimAM attention added into neck guide pay more feature related at different scales improve robustness. Experimental results show that method accurately detect complex WCI has greatly performance compared baseline.