LRS²-DM: Small Ship Target Detection in Low-Resolution Remote Sensing Images Based on Diffusion Models

With advancements in remote sensing technology, ship detection has emerged as a pivotal component in marine environmental protection and maritime traffic management. However, the significant distance of satellite imaging results in ship targets appearing as small-scale objects in the images. Current detection algorithms face challenges in accurately identifying the features of small ship targets in low-resolution settings. Therefore, this paper proposes a small ship target detection model for low-resolution remote sensing images based on diffusion models. In the first stage, cognitive conditions are used as inputs. A low-level super-resolution module enhances image clarity and facilitates the extraction of richer ship target features. The second stage employs a spatial refinement module to effectively enhance textures, edges, and other fine-grained features of small targets. Finally, an optimized loss function is designed to mitigate uncertainties arising from noise in the diffusion model for remote sensing images. Experimental results demonstrate that the proposed method achieves superior performance on the DOTA-v2.0-Ship and S-Ship datasets, attaining average precision (AP) values of 95.34% and 96.12%, respectively. Moreover, it sustains a high frame-per-second (FPS) rate, striking an optimal balance between detection accuracy and computational efficiency.