Optimised context encoder-based fusion approach with deep learning and nonlinear least square method for pan-sharpening

In this study, a hybrid optimisation strategy is used to build a deep learning system for pan sharpening. The final output image is examined using a weighted nonlinear regression model after the spatial resolution of the low resolution-hyperspectral image (LR-HIS) and high resolution multi-spectral image (HR-MSI) is increased. The deep maxout network (DMN), which used residual learning to acquire its priors, is given the HR-MSI. Moreover, DMN is trained by fractional competitive multi-verse feedback tree algorithm (FrCMVFTA). Finally, the output produced from DMN and a weighted nonlinear regression model is combined together for obtaining pan sharpened image. The PSNR value obtained by the FrCMVFTA-based DMN for the dataset Indian pines by varying the number of bands is 5.41% greater than the existing approaches. The DD value obtained by the FrCMVFTA-based DMN for the dataset Pavia by varying the number of bands is 31.47% greater than existing approaches.