Region‐of‐interest geometry optimization for shooting‐distance estimation from flash‐pulse infrared thermography of gunshot residue on textile targets

Shooting‐distance estimation from gunshot‐residue (GSR) patterns on textiles is affected by substrate properties, residue distribution, and region‐of‐interest (ROI) definition. This study evaluated how ROI geometry influences muzzle‐to‐target distance regression from flash‐pulse infrared thermography of GSR‐related thermal patterns. The analysis used 144 five‐second medium‐wave infrared recordings from four 9 19 mm semi‐automatic pistols fired at cotton textile targets at distances of 20–60 cm. Each sequence was perspective‐normalized, represented by fast Fourier transform amplitude/phase and thermographic signal reconstruction maps, and converted into radial, intensity, and physically motivated predictors. Candidate predictors were screened within the training portion of each grouped validation fold using mutual‐information ranking and a stability‐based confirmation filter, followed by Extra Trees regression. ROI geometry was evaluated over a grid of inner exclusion radius, outer crop radius, and ROI shape, comparing annular and area‐equivalent square ROIs under firearm‐held‐out grouped validation. The retrospective per‐firearm best annular ROI achieved a mean absolute error of 5.7 cm, while a candidate universal annular configuration selected within the same internal grid achieved 6.0 cm. The corresponding retrospective per‐firearm best error for the area‐equivalent square ROI was 5.8 cm. These findings indicate that ROI geometry should be treated as an explicit methodological parameter in thermographic GSR distance estimation. The proposed pipeline may support complementary, non‐contact pattern analysis, but requires external validation before operational forensic use.