Research And Application Of Visible And Infrared Image Fusion Sensing Algorithm Based On DCGAN

Due to the wide application of visible light cameras and infrared thermal imagers in many environmental perception systems,the fusion technology of visible and infrared images has a wide range of application values in the fields of video surveillance,autonomous driving,and so on.However,existing visible and infrared image fusion algorithms face challenges such as insufficient pairing data sources and difficulty in developing general registration algorithms.From the perspective of cross-domain image conversion,this thesis regards visible and infrared image fusion as a cross-domain image conversion problem,and proposes a visible and infrared image fusion algorithm based on deep convolution generative adversarial networks.The algorithm uses a convolutional neural network based on the codec structure as the generation network,and a two-class convolutional neural network as the discriminant network.Through the confrontation training of the generation network and the discriminant network,the generation network can learn from infrared images to visible images.The mapping relationship is constrained by the texture loss function,so that the generated image contains the texture characteristics of the input infrared image,so as to realize the fusion of visible and infrared image.This thesis then introduces the idea of multi-scale transform fusion,multi-scale discrimination and feature matching loss on the basis of the medium-resolution image fusion algorithm,and proposes a high-resolution visible and infrared image fusion algorithm.In order to prove the effectiveness of the algorithm in this thesis,corresponding visible and infrared image fusion experiments,target detection experiments and image classification experiments are carried out on the FLIR data set and VAIS data set.The experimental results show that the generation network can directly generate fusion images which are close to the output of the existing fusion algorithms from infrared images when using the algorithm proposed in this thesis,and the infrared images are available at all times and all climates,therefore,the visible and infrared image fusion algorithm proposed in this thesis has high application value.In addition,this thesis also designs and implements a ship yaw monitoring system,which can realize functions such as electronic fence configuration,ship detection and yaw warning.By applying the visible and infrared image fusion algorithm proposed in this thesis,the overall perception of the system is improved.