Lightweight synthetic aperture radar ship detection based on attention guidance fusion

Ship detection using synthetic aperture radar (SAR) is a useful technique for keeping an eye on marine activity. The current algorithms cannot strike the best possible balance between enhancing the detection capability of the network and lowering its weight. We present a lightweight network for identifying ships using SAR imagery. First, a lightweight downsampling module is designed to replace the traditional convolution for feature extraction, effectively reducing parameters and computation through lightweight convolution and enhancing feature representation through a hybrid downsampling strategy. In addition, a lightweight attention enhancement C3 (LAEC3) module is constructed to further reduce model parameters. LAEC3 incorporates attention mechanisms that learn relevant features from the channel and spatial dimensions, thereby enhancing ship representation. Moreover, an attention guidance fusion module (AGFM) is proposed as a method of improving relevant features and eliminating superfluous features during feature fusion. AGFM reduces the influence of intricate backdrops, therefore improving the effectiveness of ship detection. The efficacy of the proposed network was validated through experimentation conducted on two public datasets. The proposed network shows great utility with its lightweight structure and excellent performance.