Actual Image Data Generation and Application via Generative Adversarial Network

Using convolutional neural networks to identify aluminum mold numbers is an effective method. A data set with good generalization performance has a key role in improving the recognition rate of the network. The confrontation generation network can be generated and the real scene is almost the same False pictures. This paper collects aluminum mold number pictures and generates contour data pictures in general scenarios, and inputs these two types of pictures into CycleGAN for training, thereby generating fake pictures, and using pix2pix and consinGAN to generate fake pictures in IS indicators, Frechet is compared on the initial distance. Finally, the masktextspotter text recognition network is used to identify the mold number data set. The experimental results show that cycleGAN is efficient in generating aluminum mold number pictures.