Evolving Paradigms in Machine Vision for Quality Control: From Classical to Transformer-Based Approaches

Statistical Quality Control (SQC) is essential for maintaining product and infrastructure quality in manufacturing and other engineering fields. This paper reviews machine vision techniques used in SQC, tracing their development from traditional image processing methods like Hough Transform and Otsu's thresholding to deep learning models including Convolutional Neural Networks (CNNs) and new Transformer architectures. We look into foundational algorithms, CNN-based models (e.g., AlexNet, ResNet, YOLO, U-Net), and Transformer-based methods (e.g., ViT, DETR) for defect detection and quality assessment. We analyze industry-specific applications in automotive, electronics, textiles, and civil engineering, along with challenges like data scarcity, class imbalance, and computational limits. By combining key studies and recent innovations, this paper underscores the shift from manually crafted features to automated, data-driven feature learning. It discusses future directions, including multi-modal fusion, self-supervised learning, and edge AI for real-time quality control.

کلیدواژه ها:

Machine Vision ، Statistical Quality Control ، Defect Detection ، Deep Learning ، Transformer Architecture ، Image Processing ، Industrial Automation