Abstract: Aiming at the problems such as high missed detection and false detection rates caused by noise and complex nearshore background interference in the ship target detection task of synthetic aperture radar (SAR) images, as well as high model complexity and insufficient real-time detection performance, this paper proposes a method of Ship detection in SAR images based on entropy-guided loss weighted. Firstly, the interference degree between noise and the nearshore background is quantified by using image entropy, and the weight of the loss function is dynamically adjusted to enhance the feature learning of ship targets with severe interference. Then, taking YOLOv8n as the baseline model, an Efficient Pyramid Pooling Channel Attention mechanism (EPPCA) is embedded in its multi-scale feature fusion module to achieve adaptive weighted fusion of cross-scale feature channels; Finally, the attention of the network layer to the characteristics of ships is analyzed through the Grad-CAM technology, and the redundant layers are eliminated to reduce the complexity of the model. Experiments on the SSDD (SAR Ship Detection Dataset) dataset show that, compared with the original YOLOv8n, the number of model parameters of this method is reduced to 32.6% and the computational cost is reduced to 76.5%. The Precision, Recall, and mean average precision (mAP50 and MAP50-95) increased by 2.4%, 1%, 0.9%, and 1.5% respectively, significantly improving the detection accuracy while reducing the complexity of the model.