Research And Implementation Of Solar Cell Defect Identification

Machine vision is mainly about the simulation of human vision through computers, which can be used to test, judge, measure and identify the image. Image acquisition of the target object, image data processing and identification of the target are involved in Machine vision.In the industrial production process, compared with the traditional measurement test method, Machine vision has the biggest advantage of fast, accurate, reliable and intelligent to improve the consistency of product testing, product safety and reduce the labor intensity of workers, as well as implementation of efficient and automated management of production safety which has an irreplaceable role.As in solar cell production, most of the domestic product quality inspection of solar cells rely on human eye Visual inspection. Because of the human eye fatigue leading to product quality is not effectively guarantee. So it is necessary to rely on Visual inspection technology. For new problems that arise in real life, this article does some research on solar cell defect identification.The main work is as follows:1 In the Image preprocessing, the image will tilt and distortion due to the environment of light, location and defect of the capture devices. In order to ensure the identification with a unified standard, we must filter, correct, rotate, extract.and normalize the images.2 There are some reasons for PCA results, such as the large calculated amount for large data sets, the long time of feature extraction and analysis, the noise impact the results of the analysis stability and other factors. The first we should do wavelet transform on the image after the preprocessing, and then a fuzzy PCA method to calculate the characteristics of digital image, which can reduce image noise, making these feature vectors to be more helpful for the image reconstruction..In addition, according to the defects features of the image itself we extract the variance characteristics of image, characteristics of smoothness and round features.3 According to the pixel characteristics of the PCA and Image feature,we can use SVM to train and classify the training sets. There are a number of factors affect the support vector machine classification results, here the main choice of the kernel function to discuss:Select the appropriate kernel function can improve recognition accuracy and stability, Our main work is to construct a new kernel, The kernel function improves recognition accuracy in a certain extent and the parameters of the kernel function can be adjusted manually. Finally, through a large number of training samples, the experimental results can be successfully tested an unknown picture, it is so greatly to Improve the identification system performance.