Application Of Convolution Neural Network In Reproduced Anti-Counterfeiting Laser Image Recognition

There are kinds of products in our life, because there are full of fake products, the anti-counterfeiting method of products is very important. A kind of method with high intensity of anti-counterfeiting is encrypted laser anti-counterfeiting technology. This method hides information in the identification. Only by radiating the identification with laser can we get the hidden message, such as image, text and so on. The reproduced anti-counterfeiting laser image means the hidden image we get from the laser irradiation. Therefore, choosing the appropriate algorithm to improve the recognition rate of this kind of image is very meaningful.The result of the image recognition rate using the traditional template matching algorithm is to be improved. Currently, it is known that artificial neural network has a strong approximation ability of nonlinear function and has been used in many fields of image recognition, but is not yet applied in the field of reproduced anti-counterfeiting image recognition. This thesis puts the back propagation neural network algorithm into the problem. Specifically, after processing the image, extracting the feature and training the network we can get the image recognition rate. The result is better than the traditional template-matching algorithm, somehow, it can still be improved. Comparing to the deep learning method, the layers of back propagation neural network is not deep enough, so, its feature expression ability is worse then the network with deeper layers. Thus, this paper puts the convolution neural network algorithm into the recognition of reproduced anti-counterfeiting laser image and analyzes the influence of the size of the convolution kernel, the number of the convolution kernel and the layers of the network. Finally, this paper gets a common network model. Besides, the paper designs a simulation recognition system of reproduced anti-counterfeiting laser image.The experiment results show that the convolution neural network algorithm is very suitable for this specific image recognition. And by choosing the reasonable network model, the result is better than the back propagation network algorithm with shallow layers. And the artificial neural network algorithm is better than the traditional template matching algorithm.