The Research For Anomaly Detection Of Logo Images In The Mobile Phone Based On Convolutional Autoencoder

Logo image in the mobile phone,as a fundamental part for the appearance of a mobile phone,its surface defects will seriously affect the quality of smartphones and consumer experience.As a contactless and non-damaging automatic detection technology,machine vision has become a common technical means on anomaly detection of industrial image by virtue of its advantages of high precision,fast response speed,safety and reliability.It can also work stably in harsh production environment.In recent years,the detection technology based on machine vision has developed rapidly.The corresponding technical system has formed with a lot of mature visual algorithm softwares for commercial purpose has emerged.However,the accuracy of abnormal detection of Logo image in the mobile phone is inadequate due to the difference of incoming materials,illumination changes,image offset,abnormal samples are difficult to collect and other factors.Taking the mobile phone Logo image as the experimental research object,this paper proposes an unsupervised learning method based on convolution autoencoder to detect the anomaly of Logo image with the combination of optimization algorithm and digital image processing.The main work of this paper is as follows:Firstly,aiming to solve the problem that the quality of the acquisition image is uneven due to the instability of the light source parameters in the machine vision,this paper creatively uses the particle swarm optimization algorithm,which is applied to adjust the light source parameters of the lighting machine instead of the manual adjusting light source,and constructs the optimization system of the lighting parameters.In the Logo image dataset with different illumination quality,the consistency between some functions of no-reference image assessment and the artificial subjective visual effect is verified,and the image entropy function is finally determined as the fitness function in particle swarm optimization.Finally,the effectiveness verification of the lighting parameter optimization system based on particle swarm optimization algorithm is carried out on many different types of mobile phone products.The convolutional autoencoder is used to carry out unsupervised learning of mobile phone Logo image,and the reconstructed model can adapt to generate the template matching the image to be detected,which is used to compare with the image to be detected to realize anomaly detection.In order to improve the richness of the training image dataset and prevent the model from over-fitting,three methods of data argumentation are used to expand the scale of the training dataset,including the addition of Gaussian noise,contrast and brightness adjustment,rotation angle adjustment.The defect detection is carried out by using image post-processing methods such as threshold segmentation and mathematical morphological processing.Finally,the validity of Logo image anomaly detection system based on convolution autoencoder is verified on several large Logo image datasets.Based on Python,PyQt,TensorFlow,OpenCV,the lighting parameter optimization system and the mobile phone Logo image anomaly detection system have been created,and the automatic flow of Logo image anomaly detection has also been constructed.