An Efficient Reconstruction Framework for Remote Sensing Image with Large Thick Cloud Cover using Adaptive Deep Dilated Residual DenseNet

Due to the existence of cloud shadows and the clouds, it restricted the development of optical remote sensing information. Presently, various cloud shifting mechanisms are concentrated on regenerating the remote sensing information which is corrupted by the thin or small cloud layer or cover. Therefore, automated removal and detection techniques are needed in the complex environment which helps to prevent the important data of the image in the remote sensing. Owing to this, the resolution of the remote sensing image gets affected which fails to provide the clear representations. Thus, it losses the information and makes the system more complicated. Therefore, an improved deep learning-based reconstruction method is implemented to rebuild the lost data of the remote sensing information corrupted by the broad and large clouds. The proposed reconstruction model is developed by a Discrete Wavelet Transform with Adaptive Deep Dilated Residual DenseNet (DWT-ADDi-RD). By employing the bicubic interpolation-based down-sampling and up-sampling, the overall High Resolution (HR) images are transformed into Low-Resolution (LR) images. After that, the DWT structure helps to produce LR wavelet Sub-Bands (SBs) for LR images and HR wavelet Sub-Band (SB) for the HR images. By taking the differences between HR and LR wavelength, the residual image is produced. Training and testing phases are performed in this developed model. During the training stage, the remaining image of the entire image is trained by ADDi-RD with LR wavelet SBs as the input and the remaining image as the goal. In the testing module, the LR wavelet SBs query image is fed into ADDi-RD, which acquires the corresponding remaining image. Thus, the produced remaining image with LR wavelet SBs is applied to the Inverse Discrete Wavelet Transform (IDWT) to acquire the resultant super-resolution image. The significant goal of this proposed model is to enrich the efficiency of the ADDi-RD by optimising the parameters using a Random Value Enhanced Single Candidate Optimiser (RVE-SCO). In the end, the implemented system attains enhanced premium outcomes that are contrasted to the conventional approaches.