Sonar imaging technology has been continuously improving, leading to its widespread use in recognizing underwater targets. However, sonar images often suffer from low contrast, blurred edges, and high noise, which can make it difficult to extract target information during deep learing image feature extraction. This can result in the loss of target features and ultimately affect recognition accuracy. To address the issue at hand, we propose the addition of dynamic ODConv to the original yolov7 model. This will help tackle the problems of error and omission detection wuring complex background extraction and target feature loss in the feature recognition process of sonar images.By incorporating a channel attention mechanism and activation function that are more attuned to spatial features, the feature extraction and target recognition process can avoid the issue of target feature loss, ultimately leading to improved recognition accuracy.