Complete Research Dataset for SAR Superpixel Segmentation: Differentiable Superpixel Generation with Complexity-Aware Initialization and Edge Reconstruction

This dataset contains the complete research data for the study on differentiable superpixel generation for SAR imagery, including: 1. SIMULATED SAR DATA (Sim-1, Sim-2): - Full-resolution synthetic aperture radar images (512×512 pixels) - Generated using multiplicative Nakagami distribution (variance = 0.1) - Multi-texture scenarios: 3-class (cross, vertical, diagonal) and 4-class intersecting boundaries - Ground truth labels with boundary annotations 2. REAL SAR DATA (Real-1): - Ku-band China-lake image (475×446 pixels, 3m resolution) - Manual annotations for superpixel evaluation - Metadata including acquisition parameters 3. QUANTITATIVE RESULTS: - Superpixel metrics: Boundary Recall (BR), Undersegmentation Error (UE), Compactness (CO), Achievable Segmentation Accuracy (ASA) - Downstream segmentation results: Overall Accuracy (OA), F1-score, Kappa coefficient - Computational analysis: Runtime benchmarks and memory usage - Ablation study data EXCEPTIONS: Third-party benchmark datasets (Real-2: WHU-OPT-SAR and Real-3: Munich Sentinel-1B) are not included as they are publicly available from their original publications. Please refer to the README.txt for detailed citations. DATA FORMATS: - Images: GeoTIFF (.tif) and PNG (.png) - Tabular data: CSV (.csv) with UTF-8 encoding - Documentation: README.txt with complete variable definitions USAGE: This dataset supports reproducibility of the methods and results presented in the associated manuscript submitted to Remote Sensing (MDPI). All data are released under CC BY 4.0 license.