Performance Comparison of RIQE with Existing SAR GIQE Models
27th International Radar Symposium, IRS 2026, Krakow, Polonya, 19 - 21 Mayıs 2026, ss.220-225, (Tam Metin Bildiri)
- Yayın Türü: Bildiri / Tam Metin Bildiri
- Doi Numarası: 10.23919/irs70539.2026.11549102
- Basıldığı Şehir: Krakow
- Basıldığı Ülke: Polonya
- Sayfa Sayıları: ss.220-225
- Anahtar Kelimeler: GIQE, Image Quality, NIIRS, SAR, SAR Image Interpretability
- Yıldız Teknik Üniversitesi Adresli: Evet
Özet
The accurate interpretation of synthetic aperture radar (SAR) images is crucial for mission planning and sensor design in both military and civilian applications. In common practice, image interpretation is standardized using the National Image Interpretability Rating Scale (NIIRS), while NIIRS levels are estimated through General Image Quality Equations (GIQE) using appropriate imaging parameters. Most of radar GIQE formulations in the literature use resolution and noise parameters, ignore image sharpness effects, and are derived from limited data sets. This study conducts a comprehensive performance comparison between the recently introduced Radar Image Quality Equation (RIQE) and existing radar GIQE models. The introduced RIQE differs from previous GIQEs by including the Relative Edge Response (RER) parameter to represent image sharpness. RIQE framework integrates three fundamental parameters-ground sample distance (GSD), signal-to-noise ratio (SNR), and relative edge response -to enhance the prediction of SAR image interpretability. The coefficients of all models are re-estimated using a common large-scale synthetic SAR dataset to ensure statistical consistency in coefficient estimation. Model accuracy and reliability are evaluated through multiple complementary metrics, including quadratic weighted Cohen's κ, Spearman's ρ, global and in-class RMSE and ROC-AUC. Experimental results demonstrate that the RIQE formulations achieve the highest goodness of fit (κ=0.984), correlation (ρ=0.979), and discrimination capability (ROC-AUC =0.998), while maintaining stable interpretability mapping across all NIIRS levels. In addition to synthetic benchmarking, the models were further evaluated on a real SAR dataset consisting of 30 X-band Capella open-data images. Using the same complementary metrics, the proposed RIQE yields markedly lower prediction error (RMSE =0.277 NIIRS) and near-zero bias (-0.047 NIIRS) while preserving strong rank consistency (ρ=0.817). These findings confirm the robustness of the RIQE framework and provide a quantitative benchmark for assessing radar GIQE performance under consistent data and evaluation conditions.