ANOVA-driven recursive quad-tree segmentation for precision thermal defect localization in photoelectric focal plane arrays

Photoelectric Focal Plane Arrays (FPAs) are critical components in high-performance imaging systems, yet thermal defects such as localized hotspots induced by material imperfections or process variations severely degrade their temperature difference and imaging uniformity. Traditional detection methods relying on empirical thresholding and fixed-pattern noise filtering suffer from subjectivity, limited sensitivity to sub-noise-level temperature differences, and poor boundary localization. To address these challenges, this paper proposes a novel ANOVA-driven recursive quad-tree segmentation algorithm for automated and precise thermal defect localization. The method begins by acquiring a thermal matrix of the FPA under operational conditions, followed by recursively partitioning the matrix into quadrants until reaching minimum units. At each partition level, variance analysis (ANOVA) calculates inter-quadrant and intra-quadrant mean square deviations, leveraging F-statistics to determine region homogeneity. Statistically heterogeneous regions undergo further LSD (Least Significant Difference)-based multiple comparisons to identify specific defective sub-regions, while homogeneous areas terminate recursion. This hybrid approach dynamically adjusts detection granularity, enabling pixel-level boundary resolution (±3 pixels) and sensitivity to temperature differences as low as 1.5 times the system noise floor. Experimental validation on FPAs demonstrates a 4× efficiency improvement over conventional methods, achieving 256×256 matrix processing within 0.8 seconds. The algorithm eliminates human bias in threshold selection, resolves irregularly shaped hotspots, and provides quantifiable defect area metrics—advancements critical for enhancing infrared detector quality control and failure analysis. By integrating statistical rigor with multi-scale spatial analysis, this work establishes a new paradigm for defect detection in advanced photoelectric devices, with potential extensions to semiconductor wafer inspection and terahertz imaging systems.