| Abstrakt: |
It is impossible for a single photograph to convey all of the information that is available in the scene. For the purpose of producing an instructive picture, several image modalities were mixed. Because it is able to maintain a significant amount of information from its source photos, infrared and visible image fusion has garnered more interest over the last several decades. The tiny features that are available in the source pictures may be enhanced in a flexible manner. Multiscale transform techniques can give this functionality. The sparse representation approach uses less memory space than other methods, and it maintains a greater amount of information from their respective source pictures. Reducing the computational complexity and ensuring that the merged pictures are congruent with the human visual system are also benefits of the neural network technique. There is a tremendous number of data produced by the rapid changes that are occurring in current technologies, and convolutional neural networks are used to compete with the enormous amount of data. The processing power of convolutional neural networks has been put to good use in a variety of applications, including object detection, picture enhancement, surveillance, and remote sensing, among others. It is possible for CNN to extract more in-depth characteristics, and the performance of fusion is outstanding. A residual network is a network that has a high number of CNN layers, which are then followed by a residual connection. This network is appropriate for IR and visible image fusion, and it can represent more information. This study analyses the deep learning algorithms in great depth, and it has a significant amount of visibility in the environment that is always changing. [ABSTRACT FROM AUTHOR] |
