Study of Enhanced Multi-spectral Remote-sensing-satellite Image Technology Based on Improved Retinex-Net

Aimed at the problem that general Retinex-Net does not apply to multi-spectral remote-sensing-satellite image enhancement, by reference to the decomposition-enhancement network architecture of general Retinex-Net, an improved Retinex-Net image enhancement model was put forward, and is used for multi-spectral remote-sensing-satellite image processing. Firstly, a multi-spectral image was converted to an IHS image space, and decomposed into intensity, hue, and saturation components, and by using a decomposition network, the intensity component was decomposed into the two parts of illumination and reflectance. Secondly, in an enhancement network, the intensity component was corrected. Finally, the three components were fused and converted to the original multichannel color space. Experiments indicated that the improved algorithm in this paper produced a certain enhanced effect on multi-spectral remote-sensing-satellite images, but was affected by such factors as insufficient coverage types and scenarios of training sample sets. Its usability is still limited, and further experiments and researches still need to be carried out.