Research And Implementation Of Classification On The Laser Engraving Topography Defects Of Mobile Shell

Laser engraving technology is one of the most popular and efficient methods to improve the electrical resistance of electrical connection devices.With the development of mobile Internet,this technology has been applied more and more widely on the conductive surface of mobile phone shells.Resistance stability is the key to the mobile phone's ability to receive signals normally,therefore a simple and efficient surface resistance stability detection method is of great significance in the industrial field.Traditional detection methods require manual features,which has certain limitations.The neural network embeds feature selection into the hidden layer,and makes the model automatically search for expressive features through supervised learning,which has better classification performance,but requires a large number of data sets,and the neural network contains a large number of parameters which leads to low detection efficiency.In this paper,researches on the above-mentioned problems are carried out,a small image data set of laser engraving in a mobile phone shell is established,and two detection algorithms based on traditional and deep learning are designed.The main research contents here are as follows:(1)Aiming at the problem of lack of data set,this paper uses a metallurgical microscope to photograph the conductive positions in the mobile phone shell to establish a laser engraving microscope image data set.Consider the problem that the acquired data set is small,which is not conducive to network training,the data set is expanded by random rotation,histogram equalization and Laplace transform.(2)Based on the machine learning method,a defect classification method based on local binary features and gray-level co-occurrence matrices as feature extractors and support vector machines as classifiers was designed,achieving accuracy rates of 83.6%and 78.3%,respectively.(3)A model based on DenseNet121 was designed to defect laser engraving microscopic image.Aiming at the problem that the expanded laser engraving image data set is small and may not be able to converge during training,the weight of the pretrained model is loaded through the transfer learning method,and the fine-tuning strategy is used to improve the feature extraction ability of the network.Compared with the traditional classification method,the accuracy rate of 96.72% is achieved.In order to improve the detection efficiency,a multi-threading and graphics acceleration method is designed to increase the detection efficiency by more than four times.