基于边缘检测的生产线手机膜缺陷识别方法

The production and sales volume of mobile phone protective film in China is huge, but the defect inspection of mobile phone film production line still adopts the visual inspection method, so that the detection efficiency and accuracy are low. Four key problems in the detection of mobile phone protective film defects on the production line are studied: normal and defect category imbalance, high signal-to-noise ratio denoising, edge extra and defect sample detection efficiency. RST and image cutting methods are used to extend defect samples, which can solve the problems of few defective samples, defect positions and forms. The adaptive wavelet threshold and new threshold function are proposed to improve the traditional wavelet threshold denoising method and obtain excellent denoising effect. In the image edge detection technology, the improved wavelet threshold denoising method and the Otsu threshold setting method are introduced to improve the edge detection performance of traditional Canny operators and realize the effective extraction of image features. The Zernike moment with rotation, translation and scale invariance is used to express the features of the edge detection results to improve the efficiency and accuracy of pattern recognition. Finally, the suport vector machine (SVM) is used to identify the Zernike moment features of the edges of normal and defective mobile phone membranes. The experimental result shows that the method has high accuracy and fast detection speed, and meets the defect detection requirements of mobile phone film on the production line.