Research And Application Of Real Or Fake Label Appraisal Based On Deep Learning

With the rapid development,of China's economy and the improvement of people's consumption level,the trade scale of the second-hand luxury goods market has continuously expanded in recent years.The real and fake identification in the current transaction process can only be completed by humans,which is inefficient,costly,and lacks scientific basis.It is difficult to meet the needs of efficient and safe trade.With the advancement of artificial intelligence technology,deep learning can more effectively identify the deep-level features of images and complete the task of picture classification.Therefore,based on deep learning technology,this thesis automatically classifies the real and falseness of the anti-counterfeit washing label images in luxury brand GUCCI bags,which has practical application value for completing the real verification by computer and reducing trade risks.The difference between the real of fake labels is mainly reflected in the texture materials and font printing technology.Therefore,this thesis first uses the classic edge detection algorithm to perform feature extraction and usability analysis on the labels.Then,a classifier based on the combination of directional gradient histogram features and support vector machine algorithm is designed.After simulation and optimization,the classification accuracy rate reaches 90.25%,but this method is susceptible to interference from complex backgrounds and has the problem of unstable results.According to the structural characteristics of the VGGNet-16 model,this thesis designs a lightweight convolutional neural network LUX-CNN,used to classify the true and false of the small black mark.Finally,the model is further optimized by adding a spatial transformation network,attention module,and replacement classifier to the LUX-CNN as SVM.The experiment is carried out on the self-built small black mark data set.The experimental results show that the deep learning model is more suitable for the real and fake classification of small black marks.Through comparative experiments,LUX-CNN achieved higher accuracy than LeNet and AlexNet and shorter recognition time than VGGNet-16.For individual features in the data set that are not obvious,adding a spatial change network before or after the convolutional layer makes the features get better performance,and the recognition rate reaches 98.54%.Combining the attention mechanism and replacement classifier as the optimization strategy of SVM,the recognition rate is up to 99.26%.