Surface Defect Detection Based on Machine Learning and Data Augmentation

The development of industrial automation and intelligent manufacturing has promoted the research of computer vision and machine learning in the automated detection of surface defects in metal and plastic products, to solve the problems of low efficiency and high inconsistency in traditional manual detection. In response to such demandsthis paper proposes a lightweight model that can be used for high-precision industrial defect detection. Data Preprocessing First, we use Python to write programs to read images. Secondly, image features are extracted. In this paper, the HOG algorithm is used to extract the features of the image, and the principal component analysis (PCA) is used to reduce the dimension of the feature. Finally, we trained a defect target detection model based on grid division and LightGBM, which we call the Grid-HOG-PCA-LightGBM model. The recognition accuracy of the model can reach 98%, and it only takes up 4000B memory size. Our models not only enable automated inspection, increasing inspection efficiency and accuracy; they also reduce costs in the production process by enabling continuous and efficient inspection.