A Convolutional Neural Network for Segmentation of Background Texture and Defect on Copper Clad Lamination Surface

This research interprets the design and test process of copper clad lamination surface defects detection. The system was included four following stages: image acquisition, image pre-processing and segmentation, convolutional neural network design and image classification. Image processing method and pattern recognition algorithm are utilized in the system. First, the author applies the smoothing filters to eliminate noise from the images and segmenting a defect from background texture. Then, the convolutional neural network architecture is created to learn local feature of defect and background texture. Finally, defect and background images from segmentation step are collected and fed into a convolutional neural network to train and perform the classification task. The classification results demonstrate that the proposed method can re-checked false positive detect from the conventional Sobel edge detection, Hence the accuracy was increased from 78.1% to 98.2%.