Underwater target detection optimization algorithm based on YOLOv7

Research on underwater target detection is highly valuable in a variety of domains, including resource exploration, marine ecology, and the marine environment. The accuracy of detection is frequently unsatisfied due to blur distortion and dispersion in the underwater environment, particularly for small underwater target objects. To tackle such problems, this paper proposes underwater target detection method, named YOLOv7-ASF. This framework significantly increases the speed and precision of underwater target recognition by combining spatial and scale characteristics. Employing the YOLOv7 framework as the backbone, we first use the Scale Sequence Feature Fusion (SSFF) module to improve the network's multi-scale information extraction capabilities. To provide comprehensive information, we fuse feature maps of various scales by the TPE (Triple Feature Encoder) module. Moreover, to enhance the model's detection performance, the channel and position attention mechanism (CPAM), which focuses on small objects connected to information channels and spatial positions, is added to combine the SSFF and TPE modules. Experiments on the URPC2020 data set demonstrate the effectiveness of the proposed method.