Fusion of Visible and Infrared Images Using a Reinforcement Learning System Based on Fuzzy Logic and Convolution Optimized with Wild Horse Algorithm

To create a fusion image with more information, complementary data from similar images taken by multiple types of sensors are combined into a single image through the process of infrared and visible image fusion. Existing fusion approaches based on machine learning still struggle with how to better preserve the detail information in the source images. To improve and supplement the fusion image information, a hybrid reinforcement learning system based on fuzzy logic and convolutional network and filtering designed to fuse visible and infrared images is used in this work. This hybrid reinforcement learning system was optimized using algorithms including wild horse optimization (WHO), genetic algorithm (GA), and particle swarm optimization (PSO) to improve specific fusion metrics such as image correlation, similarity coefficient, image entropy, and signal-to-noise ratio. The system aims to preserve the detail information in the final image by adding thermal information from the infrared image to the visible image, which is achieved by performing a series of fusion operations on the input images, including image detail enhancement with the help of a convolutional network and meaningful image fusion with fuzzy logic semantic model and filtering operations to increase clarity. As part of the validation process, the advantages of the proposed algorithm were compared with other classical algorithms using the TNO dataset. Successfully, the proposed method has been able to increase the SSIM parameters to 1.8594 and PSNR to 61.42. Based on the experimental results, our proposed method outperforms previous fusion methods in terms of subjective and objective evaluations.