Research On Infrared And Visible Images Fusion Method And Its Applications Based On Autoencoder Network

The infrared and visible image fusion technology has been widely used in security protection,military investigation,intelligent transportation and other fields.The related image fusion techniques have become important research contents in the field of image processing.In recent years,infrared and visible image fusion methods have been developed rapidly,but there are still problems such as loss of source image structure information,underutilization of deeplevel feature information,insufficient preservation of infrared target detail information,and low contrast of fusion results.The infrared and visible image fusion methods based on autoencoder network are discussed to improve the performance and expand their applications in this dissertation.(1)An improved infrared and visible image fusion method based on residual dense network and gradient operation is given by introducing a residual dense network with gradient operation.Specifically,a residual dense network containing gradient operations is introduced in the encoder module to make it consist of three residual dense blocks containing gradient operations in the residual stream through residual connections,which helps the encoder module fully Preserve the intermediate layer information of the source image and improve its ability to extract fine-grained information in the source image,thereby improving the ability of the entire model to extract detailed information.(2)An improved infrared and visible image fusion method based on DenseFuse and space transformer is given by introducing a fusion strategy based on space transformer.Specifically,based on the DenseFuse fusion network,the local and global information of image features are fused by introducing a spatial transformer in the fusion layer,where the spatial branch uses convection blocks and bottleneck layers to capture local features,and the transformer branch uses self-attention The mechanism models long-distance dependencies and learns global contextual features based on an axial attention mechanism.A fused feature map containing enhanced local features and global context information is obtained by superimposing local features and global context features.(3)The improved method of infrared and visible image fusion based on autoencoder network is applied to the TNO dataset and compared with the existing classical methods.The experimental results show that the improved method given in this paper can effectively realize the fusion task of infrared and visible light images,and the obtained fusion image can fully display the saliency information of infrared images and the detailed texture information of visible light images,and achieved a better fusion performance.The performance of fusing infrared and visible images is improved based on the selfencoder network to a certain extent,and the fusion images with high quality are obtained in this dissertation.The work can be applied to many fields such as scene recognition,equipment monitoring,ship detection and other fields to improve the performance.