Underwater acoustic target detection is a task that presents both challenges and application potential. With the successful application of deep learning technology in computer vision and signal processing, an increasing number of researchers are applying these techniques and concepts to the field of underwater acoustic target detection, resulting in significant progress. However, existing methods still have room for performance improvement, as they often overlook the unique characteristics of underwater tasks. To address the challenges of high noise, poor image quality, and low target recognition accuracy in forward-looking sonar (FLS) images, This paper introduces a network designed to detect underwater acoustic targets, which is built on the principle of comprehensive feature extraction. By combining the convolution operation with the self-attention mechanism, this paper designed an enhanced version of the Comprehensive Feature Extraction (CFE) module. The convolution branch is a core component of the network that operates on the input image by sliding the convolution kernel to efficiently capture local features such as edges, textures, corners, and other relevant patterns, while the self-attention mechanism is designed to capture global patterns. We integrate the CFE module into the backbone of the baseline Yolov10 network and conduct experiments on two FLS image datasets. Experimental results show that our proposed model exhibits excellent detection performance in object detection tasks and demonstrates strong competitiveness in the field, objectively demonstrating the advantages of the new model.