Machine vision intelligence for product defect inspection based on deep learning and Hough transform

Abstract Machine vision based product inspection methods have been widely investigated to improve product quality and reduce labour costs. Recent advancement in deep learning provides advanced analytics tools with high inspection accuracy and robustness. However, the construction of deep learning model is typically computationally expensive, which may not match the requirements for quick inspection. Therefore, this paper presents a new deep learning based machine vision inspection method to identify and classify defective product without loss of accuracy. In specific, Gaussian filter is first performed on the acquired image to limit random noise. Then, a region of interest (ROI) extracting project is conducted based on Hough transform to remove the unrelated background, thereby offloading the computational burden of the subsequent identification process. The construction of the identification module is based on convolutional neural network, whereas inverted residual block is introduced as the basic block to strike a good balance between identification accuracy and computational efficiency. The superior inspection performance is obtained using the proposed method with a large amount of dataset which consists of defective and defect-free bottle images.