1. AFSar is an anchor-free SAR target detection algorithm based on multiscale enhancement representation learning.
2. The algorithm utilizes the YOLOX architecture as the basic framework and incorporates a lightweight backbone (MobileNetV2S) to improve computational efficiency and multiscale feature extraction.
3. A novel attention enhancement PAN module called CSEMPAN is proposed, which highlights the unique strong scattering characteristics of SAR targets by integrating channel and spatial attention mechanisms. Additionally, a target detection head called ESPHead is introduced to enhance the model's ability to detect targets with different scales using dilated convolution with varying dilated rates.