Ship target detection in SAR images based on lightweight SNC-HL model

Target detection is a crucial aspect of radar image interpretation. The combination of SAR technology and deep learning network models has become a significant approach for ship target detection. Compared to traditional methods, deep learning offers powerful data processing and feature extraction capabilities. However, increasing the width and depth of convolutional neural networks can lead to higher computational complexity and memory consumption, which is not conducive to deployment on terminal devices. To counter these obstacles, this paper proposes a lightweight model SNC-LH for detecting ship targets in sea clutter backgrounds. First, the original convolution and fusion bottleneck modules (C2f) in the network are replaced with a lightweight SNCBlock to strengthen the network's feature extraction performance. Second, the decoupled detection head is replaced with a lightweight detection head (LH) to reduce model complexity, minimize parameter adjustment costs, and improve detection accuracy. Finally, extensive experiments are conducted on the SSDD datasets. The experimental results demonstrate that compared to the original algorithm, the SNC-LH model achieves higher detection accuracy while reducing the number of parameters and GFLOPs to 23.5% and 3 of the original, respectively, meeting the real-time and resource-efficiency requirements of the application scenario.