(1.College of Physics and Information Engineering，Fuzhou University，Fuzhou 350116，China； 2.Fujian Science & Technology Innovation Laboratory for Optoelectronic Information of China，Fuzhou 350116，China)
Abstract： In view of the problems of anti-counterfeiting technology, such as cloneable, low reliability, and high cost, this paper proposed an identification method for random wrinkle formed by compressive stress caused by the mismatch of thermal expansion index. The paper used DeepLabv3, a edge of deep convolution network classification algorithm, for classification and recognition. Through optimizing the full connectivity layer and setting different neuron nodes, the classification accuracy of recognition network was improved, the training time was reduced, the training accuracy rate was as high as 96.58%, the network model for accurate recognition of wrinkle texture pattern was acquired, and the security purpose of anti-counterfeiting was realized.
Key words : anti-counterfeiting；deep learning；DeepLabv3；image classification Artificial Intelligence