Real-Time Detection of Surface Floating Garbage Based on Improved YOLOv7.

Cleaning up surface floating garbage is of great significance to improve the ecological environment of the waters, of which the detection of surface floating garbage is a crucial step. It is difficult to distinguish the background and garbage and solve the problem of small targets of the existing target detection methods. Therefore, an improved YOLOv7 network is proposed to achieve efficient and high-precision detection of surface floating garbage. The newly proposed DS method is applied to original YOLOv7 network for multi-scale feature fusion. Meanwhile, CBAM attention mechanism module is applied to network backbone to improve detection precision. The public data set FLOW is used to validate the method in this paper. The simulation test results show that the proposed method significantly improves the detection precision without reducing too much detection speed, which proves the effectiveness of the proposed method. Finally, the proposed method is deployed to an NVIDIA Jetson device and applied to real-time surface floating garbage detection, which proves its practicability.