一种基于轻量级深度网络的无参考光学遥感图像增强算法

A Non-reference Optical Remote Sensing Image Enhancement Algorithm Based on Lightweight Deep Network

光学遥感图像由于传感器性能下降,恶劣天气等因素的影响,导致图像品质变差。表现为图像对比度不高、细节不清晰、颜色失真、噪声大等问题。文章提出一种无参考光学遥感图像增强算法,旨在训练一个轻量级的深度网络,以估计像素和高阶曲线的动态范围调整给定的图像。在网络训练过程中不需要任何成对或不成对的数据,而是通过一组无参考损失函数隐含地测量了图像增强品质并驱动了网络的学习。实验结果表明,文章算法的性能在多种指标上和视觉上均取得优异效果,能够适应更多的光照条件。

Optical remote sensing images are affected by factors such as sensor performance degradation and bad weather,resulting in poor image quality.It is manifested as problems such as low image contrast,unclear details,color distortion,and large noise.This paper proposes a non-reference optical remote sensing image enhancement algorithm that aims to train a lightweight deep network to estimate the dynamic range of pixels and higher-order curves to adjust a given image.No paired or unpaired data is required during network training,instead the enhancement quality is implicitly measured and drives the learning of the network through a set of no-reference loss functions.The experimental results show that the performance of the proposed algorithm has achieved excellent results in various indicators and visuals,and can adapt to more lighting conditions.