Design and Implementation of an Infrared Image Generative Model

Night vision technology provides an important means for night observation by using infrared radiation of the target as the imaging basis. Recently, artificial neural network (ANN) has become a research hotspot. Realizing the classification and recognition of infrared images by ANN is helpful to the intelligent development of night vision technology, as well as improve the accurate judgment of the surrounding environment by unmanned vehicles at night, and better ensure the safety of people and vehicles. However, the existing infrared image data set contains a small number of images, and the types of scenes and targets presented in the images are single, which cannot support the training of ANN. ANN cannot obtain the classification function of infrared images through full learning. In this paper, an infrared image generative model based on generative adversarial network (GAN) is proposed, which can convert the existing visible light data set into the infrared version of the data set in a short time, and then solve the problem of lack of infrared image data sets. To train ResNet by infrared version data set can realize its classification function for infrared images. The model with the existing image generative model is compared through several evaluation indexes. The results show that the model is more effective in generative infrared images. Inputting ImageNet data set into this model can generate ImageNet infrared version data set. ResNet trained by this data set has a classification and recognition accuracy of 91.28%, which is higher than the experimental results of existing image generative models. The design and implementation of infrared image generative model in this paper promote the application of artificial intelligence and depth learning in night vision technology and other fields.