To address the challenges of small target omission and dense occlusion interference in ship detection from high-resolution optical remote sensing images, this paper proposes a dense ship target detection algorithm based on global scene understanding and local detail perception. First, to enhance the backbone network’s capability in extracting features of small targets, a fine-grained perception module (FGPM) is designed. Its multi-scale parallel feature extraction strategy and dual-attention enhancement mechanism effectively alleviate the inherent information loss of small targets while improving the signal-to-noise ratio between the foreground and the complex background; Second, a multi-scale global and local (MGLF) joint Feature fusion module is designed to synergistically optimize global scene understanding and local detail perception, enhancing the model’s robustness and adaptive capability in complex regions; Finally, ADown downsampling is introduced to replace traditional convolutions, preventing excessive loss of contextual information during the feature map size reduction process. Experimental results on the DIOR and NWPU datasets demonstrate that the proposed algorithm achieves improvements of 1% to 5% in precision, recall, and mean average precision compared to the baseline method, effectively validating its ability to ensure detection accuracy while reducing false detections and missed detections in optical remote sensing-based dense ship detection.