Defect Recognition in Array Chips: A Two-Dimensional Otsu-SVM Method Optimized by PSO++

Accurate segmentation and classification of array chip defects are crucial for intelligent chip manufacturing. To address issues of inaccurate threshold segmentation and redundant high-dimensional features, this paper proposes an array chip defect recognition method based on an improved Particle Swarm Optimization (PSO++). PSO + + is combined with 2D Otsu for plastic encapsulation defect segmentation, dynamically adjusting inertia weights and acceleration coefficients to achieve optimal thresholds efficiently. For pin defects, a contour-rectangularity-based algorithm precisely locates defect regions. PSO + + further selects optimal feature subsets for an SVM classifier, improving accuracy while reducing dimensionality. Experiments show PSO++-Otsu shortens segmentation time by 40–49%, and PSO++-SVM achieves 96.20–98.60% accuracy, outperforming existing methods by 2.20–7.47%.