Small Visible Defect Detection of Small Sample based on the Fusion of Multiple Features

Small visible defects on the surfaces of products affect their appearance and sales. Machine vision is a common and effective detection method for detecting the defects. At present, visual detection based on deep learning has made a great breakthrough in terms of accuracy. However, this method is not suitable for the production of small-scaled products since it relies on a large amount of data to become effective and the training of the deep learning model usually requires a very long time. Typically, the production of a new design is limited to a small scale, and the initial sample is not sufficient for the evaluation function to predict small visible defects. To address this issue, a fast approach for small visible defect detection is developed based on a fusion of multiple features using a small sample of data, also a recursive scheme is developed to tune the coefficients of the evaluation function as the sample size increases during the production process. To validate the proposed approach, a set of experiments using a patch of 500 watch dials and 500 jewelry accessories were conducted. It is shown that this method can effectively improve the prediction accuracy of the evaluation function. © 2022 CAD Solutions, LLC,.