HOG-PCA-Block-Voting: Lighter and Smaller but Faster

This study proposes a novel defect detection framework utilizing Histogram of Oriented Gradients (HOG) for feature extraction and Principal Component Analysis (PCA) for dimensionality reduction, culminating in a lightweight, yet efficient, HOG-PCA-Block Voting model. Aimed at enhancing defect detection in manufacturing, especially on mobile and handheld devices, this research addresses the critical need for models that balance computational resource constraints with high detection accuracy. Through extensive experimentation, the proposed model demonstrates superior performance in accurately identifying surface defects, supported by a robust evaluation system encompassing accuracy, precision, recall, runtime, and size metrics. Key findings reveal the Voting classifier, an ensemble of Logistic Regression, KNN, Naive Bayes, SVM, and Decision Tree models, significantly improves detection precision and operational efficiency. The application of image block techniques for defect localization further evidences the model's adaptability and precision. Despite its advantages, the study acknowledges limitations, such as dataset comprehensiveness, and suggests future research avenues, including deep learning integration and advanced ensemble methods, to enhance defect detection capabilities. The HOG-PCA-Block Voting model, therefore, stands as a significant advancement in automated quality control, offering a practical, efficient tool for real-world manufacturing scenarios.