Surface-detection Algorithm For Components

In recent years, with using the surface mounted devices more broadly and theaccelerating process of terminal machine processing industry transferred to China, China hasbecome the world’s important consumer market of surface mount devices, bringing the hugedevelopment space to the domestic electronic components enterprises. In order to ensure thequality of surface mount devices, form factor and surface defect of surface mount devicesmust be detected.In this paper, considering SMD (Surface Mounted Devices) visual inspection equipmentof surface mount devices as breakthrough point, taking Chip capacitors, Sheet resistance andChip Inductors as main study object, defects and damage types occurred after probing areanalyzed. Number of bright spots, dark points, area, blocks and the edge points which areposition-independent and image features-independent are collected. According to imagefeatures and testing processes, six types of defects are divided. Because there is no relationbetween image features and components position, so defect inspection of this paper requiresno location of surface mounted devices and link of image guide.In order to achieve the extraction of target components and eliminate interference image,through the binarization processing and analysis of blob, image contained the chip isintercepted and then new image is processed. To solve the problem of automaticthreshold-selecting image segmentation, threshold-search method is presented to confirmthreshold value.This paper uses method of morphology to detect the defects in different regions of thechip. To the scratches and cracks which are hard to detect, edge detection-edge point analysisis adopted. There are three algorithms of edge detection to compare with each other. The firstalgorithm is the multi-scale weighted edge detection algorithm based on Canny operator. Itbrings in multi-scale thought of wavelet and uses three different scales of the Gaussian filter.At each scale, the image is filtered and gradient is calculated, and then these gradient mapsare weighted to get the final synthesis.Since the computation of this method is large, it is difficult to meet the needs of actualtesting. The multi-scale B-spline wavelet edge detection is put to use. Time complexity ofB-spline wavelet edge detection is not high. And the appearance of the image contourextraction is accurate. But it’s too sensitive for internal texture. Therefore, adaptive thresholdSUSAN algorithm is used to extract the image edge. Experiments demonstrate SUSANalgorithm is able to extract scratches and cracks of components surface more stably. Using detection algorithm proposed in this paper, a large number of defect componentimage are detected. The recognition accuracy rate reaches to above97%.