Haze Removal for Remote Sensing Images Using Residual Attentive Atmospheric Scattering Network
Xianghua Niu, Wei Huang, Rui Huang · 2022 IEEE 24th International Workshop on Multimedia Signal Processing (MMSP) · 2022
Due to the complexity of remote sensing image scene, it is a great challenge to obtain high-quality haze-free remote sensing images. In this paper, a haze removal network for remote sensing images called residual attentive atmospheric scattering network (RAASNet) is proposed. The network first predicts transmission map and atmospheric light value combined with neural network, and calculates preliminary haze removal result through the atmospheric scattering model. Then a haze-free image is obtained through the enhancement model. In this network, residual attentive block is added into the transmission map prediction network to enhance accuracy of the predicted image, and adjustment loss function makes the output closer to real haze-free image. Through qualitative and quantitative analysis of experimental results compared with other state-of-the-art methods, our method has good performance in haze removal for remote sensing images with different landscape and different resolution. And outputs of our network have high color fidelity and rich image details.