Improved Remote Sensing Image Registration of Residual Block Densely Connected Network Based on Reinforcement Learning
Ying Chen, Chen Lei, Qi Zhang, Wei Wang, Jiahao Wang · 2021
As one of the vital research in terms of image processing, NIR (NIR) and RGB (RGB) remote sensing image registration (RSIR) has attracted the attention of many scholars. We proposed the densely connected neural network improved by residual blocks (RBDCNN) to extract features to generate Q values, which improves the performance of registration, and uses the distance difference between the transformation matrix of the reference image and the float image. Compared with the existing method, our registration result is closer to the reference image. Our method was tested on five data sets in Landsat8, and the simulation results show that our network effectively improved the recognition of NIR and RGB remote sensing images and the accuracy of registration.