<p>Remote sensing small object images exhibit characteristics such as blurred boundaries, complex target shapes, and dense distributions. Existing satellite remote sensing technologies enable real-time observation of ground objects; however, they often suffer from missed detections, false detections, and inaccurate segmentation. To address these issues, we propose a Phase Enhancement and Optimize Sorting-based remote sensing small object instance segmentation algorithm, named RELACT. We design a Phase Enhancement Matching Module, which magnifies image textures and establishes precise associations to effectively achieve hierarchical extraction of texture-aware features. To enhance the network's ability to detect small objects in remote sensing images, we develop a Reverse Boundary Weighted Attention Network. This network performs reverse weighting on the boundaries of remote sensing images and incorporates a multi-scale feature enhancement module, effectively facilitating the high-quality fusion of boundary region segmentation mask information and image target features. Finally, we introduce BRS Loss to optimize the sorting of positive and negative sample scores, effectively addressing the issue of class imbalance between remote sensing small objects and background. We conduct extensive ablation and comparative experiments on three challenging benchmark datasets: iSAID-100, NWPU VHR-10, and HRSID, to evaluate the performance of the proposed RELACT model. The experimental results demonstrate that RELACT achieves AP scores of 38.7%, 65.3%, and 57.7% on the iSAID-100, NWPU VHR-10, and HRSID datasets, respectively, effectively enabling precise recognition and high-quality segmentation of multi-scale and complex objects in remote sensing imagery. The code is available at https://github.com/aokseen/RELACT.</p>