ISAR Image Geometric Standardization via Bipartite Convex Hull Affine Transformation Reconstruction for Automatic Ship Target Recognition
Xinfei Jin, Zihan Xu, Xinbo Xu, Hongxu Li, Fulin Su · IEEE Transactions on Aerospace and Electronic Systems · 2025
Inverse Synthetic Aperture Radar (ISAR) images are extensively utilized in Radar Automatic Target Recognition (RATR). However, the target motion and the variations in the Line of Sight (LoS) significantly impact ISAR images. These factors lead to attitude sensitivity, which decreases recognition accuracy. To overcome this challenge, this paper presents an algorithm named Bipartite Convex Hull Affine Transformation Reconstruction (BCHATR) to assist with ISAR image geometric standardization for automatic target recognition. First, the geometric features such as the convex hull, centerline, and centroid are extracted. Next, the convex hull vertices are filtered to remove closely located points. Following that, the extracted centerline divides the convex hull into two parts. Subsequently, for the upper part, the vertex farthest from the centerline is selected as a feature point. For the lower part, the vertex corresponding to the maximum Intersection over Union (IoU) is selected as a feature point. The feature points, involving the selected vertices, the centerline endpoints, and the centroid, are all demonstrated to be affine invariant and are used to calculate the affine transformation coefficients. Following the affine transformation, the ISAR images are standardized with reduced attitude sensitivity. Finally, the effectiveness of BCHATR is further validated through recognition experiments using standardized ISAR images as templates. Experiments based on both simulated and measured data confirm the superior performance of the proposed method, which achieves the highest mean accuracy among all compared methods.