An Iterative Polygon-Based Approach for Accurate Ground Object Extraction from Synthetic Remote Sensing Imagery

This study proposes an efficient ground object extraction algorithm based on iterative polygon fitting, aimed at addressing the challenges of accurately segmenting objects with complex geometric boundaries in remote sensing imagery. Using a synthetic image composed of distinct ground object types, the algorithm’s iterative convergence behavior and fitting capability were thoroughly evaluated. The method integrates multiple proposal moves — including polygon addition, deletion, merging, and node adjustment — to flexibly adapt to object shape variations. Experimental results demonstrate that the algorithm achieves stable convergence and precise object delineation after an optimal number of iterations. Comparative analysis with conventional pixel-based statistical methods, regular marked point process models, and active contour approaches shows that the proposed algorithm significantly outperforms its counterparts, particularly in handling irregular object boundaries and reducing segmentation noise. Quantitative metrics, including user accuracy, kappa coefficient, and F1 score, further validate the superiority of the proposed method in both segmentation completeness and geometric precision, demonstrating its potential for practical applications in object extraction tasks involving heterogeneous and complex surface features.