A Deep Multitask Convolutional Neural Network for Remote Sensing Image Super-Resolution and Colorization

Remote sensing data have become increasingly vital in target detection, disaster monitoring, and military surveillance. Abundant pan-sharpening super-resolution (SR) methods based on deep learning been proposed achieved remarkable performance. However, requires paired panchromatic (PAN) multispectral (MS) images, SR cannot increase the spectral resolution of PAN. Thus, we introduce a computational imaging-based method to recover or produce incomplete single PAN MS. This work also explores integration multiple tasks by neural network. We start with colorization, study feasibility simultaneously finishing use model trained colorization finish without A generic network, remote image improvement network (RSI-Net), is designed for SR, simultaneous pan-sharpening. To verify its performance, RSI-Net compared state-of-the-art methods. Experiments show that can be competitive visual effects evaluation indexes, it performs well at finishes only needs input Our experiments confirm effect integrating tasks.