Research On Surface Line Defect Detection Of High-density Flexible Integrated Circuit Substrate Based On Superpixel Segmentation

Flexible Integrated Circuit Substrate(FICS)is a bare circuit board of unpackaged electronic devices,and as one of the core components of electronic products,its quality inspection is very significant.With electronic products becoming thinner and lighter,the wiring of circuit boards has become more and more integrated,thus the quality inspection method has gradually changed from manual visual inspection to machine vision-based inspection.Therefore,this paper is based on self-designed FICS visual inspection system and conducting in-depth research on the visual inspection technology of FICS surface line defect detection.The main work of this paper is as follows:(1)This paper proposes a selective refinement superpixel segmentation(Ref-SS)from the perspective of clustering segmentation.Ref-SS designs a two-stage refinement that from region level and pixel level to refine unqualified superpixels,which contain multiple objects in original superpixel segmentations.The experimental results on public Berkeley image data set BSDS500 prove that Ref-SS outperforms other superpixel algorithms,especially in boundary recall,which is averagely 7.3% higher than classical algorithm SLIC,and 5.9% higher than SEEDS algorithm,and 2% higher than FLIC algorithm.(2)To detect the internal holes and oxidation defects on the surface line of FICS,based on superpixel segmentation result of Ref-SS,this paper proposes an algorithm which combining region saliency detection and brightness feature.By extracting and analyzing the features of superpixels located on copper line to identify those superpixels containing above defects.The experiment result shows that this detection method is robust to brightness,color and texture.For FICS image with a size of 417×417,when the size of defect is larger than 60 pixels,the detection precision can reach 100%.(3)To detect the sawtooth defect on the edge of FICS line,based on superpixel segmentation result of Ref-SS,this paper designs a line width measurement algorithm that adopting piecewise fitting.First,extracting the contours of copper line based on Ref-SS superpixel segmentation result.Then,segmenting each contour with a setting step size,and according to line features to fit straight lines and classify contour's line type.Finally,calculating the average line width and maximum and minimum line width based on relevant line segments,so as to detect the sawtooth defects.Experiment analysis shows that the piecewise fitting method proposed in this paper is faster than other fitting methods such as least square fitting and Hough transform,besides it can retain the details of uneven lines.