<p>Synthetic Aperture Radar (SAR) ship object detection is critical for ocean management and national security. The variability of image scales and the high computational cost of real-time detection are currently two important challenges for object detection. Multi-scale feature fusion is the current mainstream solution, however, most of the feature fusion only uses summation or cascade to fuse pyramidal features, ignoring the scale features of the image itself. Current mainstream methods often involve large backbone networks with a high number of parameters, which reduces inference speed and presents challenges for deployment on satellite platforms. To address these issues, we propose an ultra-lightweight SAR ship target detection model based on multi-scale feature fusion and pruning distillation. This model includes a multi-scale feature fusion module, an enhanced feature extraction module, a channel pruning network, and a knowledge distillation network. This model aims to achieve multi-scale feature extraction and fusion while maintaining lossless compression. The multi-scale feature fusion module performs feature extraction by enlarging and combining semantic information from different scales of the image. The enhanced feature extraction module reduces redundant computation through feature division and merging, while employing layer aggregation to enhance feature extraction capabilities. The channel pruning network uses a slim structured pruning model. The Multiple Knowledge Distillation (MKD) approach leverages SAR-HP as the teacher model and its pruned version as the student model, aiming to preserve detection accuracy while reducing the model’s parameters. Experimental results on the SSDD dataset demonstrate that our model significantly improves accuracy, reduces model size by 51.6%, and increases inference speed by 48.7%.</p>