Infrared Image Time Extension Method Based On Deep Learning

High-definition infrared scene simulation techniques rely on high-quality infrared textures,which require a large amount of infrared texture data at different time conditions.Due to the limited nature of real-time texture data,it is necessary to study the infrared image period expansion technique to obtain a highly realistic infrared texture.Infrared image period expansion can solve the problem of time-consuming and labor-intensive image processing.In addition,the real-time infrared image can present the current time temperature and energy distribution of the scene,with a true infrared texture distribution.Therefore,based on real-time infrared images,this paper studies the generation of high-realistic infrared texture images at different time periods.Based on the analysis of the current development status of infrared image simulation,this paper introduces the artificial neural network into the time period of infrared image.The main research content are:(1)The application of deep learning method in the field of time expansion of infrared image is conducted exploratory research.By using the current mature deep learning algorithm and adjusting to the optimized parameters,the infrared image at one moment will become the image of another time,and an infrared image is generated.At the same time,the reverse conversion image is performed.The resulting poor quality images are optimized in the inverse expansion results.(2)This paper makes an exploratory study on the different radiance of the scene caused by the different directions of the sun.The morning and afternoon images are used for network training.The transformation of the two states of the back-shade and the sun-facing is completed.This paper makes an exploratory prediction of the trend of the shadow.(3)The extended image generated by deep learning is optimized and expanded.In this paper,based on the principle that temperature and gray scale show a linear relationship within a certain range,the infrared image generated by deep learning is extended to other time periods using gray scale transformation.