<![CDATA[This study develops a visual welding defect detection system based on Raspberry Pi by utilizing Gray Level Co-occurrence Matrix (GLCM), Histogram of Oriented Gradients (HOG), Principal Component Analysis (PCA), and Support Vector Machine (SVM) methods to identify various types of welding defects, including porosity, undercut, burn-through, overlap, and spatter. The system is designed to operate automatically and in real-time through several processing stages, including image acquisition, preprocessing to enhance image quality, Region of Interest (ROI) segmentation, feature extraction of texture and shape, dimensionality reduction using PCA, and multiclass classification using SVM. In addition, this study aims to evaluate the effect of image acquisition conditions on system performance, particularly variations in lighting, distance, and camera angle, which are critical factors in industrial implementation. Experiments were conducted under several scenarios to determine the optimal parameters that yield the best performance. The results show that the optimal condition is achieved at a lighting level of 50 lux, a camera distance of 10 cm, and a viewing angle of 20°. Under these conditions, the system achieves an accuracy of 100% for normal part classification and 94.4% for multiclass classification. The precision and recall values both reach 94%, with an F1-score of 93%, indicating a balanced performance in detecting different types of welding defects. Overall, the results demonstrate that the proposed system has strong potential as an effective, efficient, and real-time automated inspection solution for welding quality in industrial manufacturing environments.]]>
