基于Retinex和多注意力的低光照航拍图像增强方法

Low-Light Aerial Image Enhancement Method Based on Retinex and Multi-Attention Mechanism

针对低光照航拍图像亮度低、对比度弱、噪声多、细节缺失等问题,提出一种基于Retinex和多注意力机制的低光照航拍图像增强(MARNet)方法。首先,将低光照航拍图像分解为光照图和反射图,再将CBAM注意力机制引入噪声调整网络,让网络更加关注高噪区域,去除反射图中大量噪声;然后,设计了由上下采样结构组成的光照调整网络,引入通道注意力机制,提升光照图亮度,同时,加入区域损失函数,提高细节对比度;最后,为实现低光照近地面目标检测与跟踪,利用低光照图像合成方法,加入真实噪声,制作了一套低光照航拍配对数据集。实验结果表明,所提方法在提高图像亮度、减少噪声的同时还原了细节信息,3项性能指标PSNR,SSIM和NIQE及人类视觉感知效果均有所提升。

Aiming at the problems of low brightness,weak contrast,high noise,and lack of details in low-light aerial images,a low-light aerial image enhancement algorithm(MARNet)based on Retinex theory and multi-attention mechanism is proposed.Firstly,the low-light aerial image is decomposed into light map and reflection map,and the CBAM attention mechanism is introduced into the noise adjustment network,so that the network pays more attention to the high-noise area and removes lots of the noises in reflection map.Secondly,a lighting adjustment network is designed by using up-and down-sampling structure.The channel attention mechanism is introduced to improve the brightness of the light map,and the regional loss function is added to improve the detail contrast.Finally,a set of low-light aerial images dataset is established by using the low-light image synthesis method with real noises.The experimental results show that the proposed method can enhance brightnes,reduce noise and restore details,and PSNR,SSIM,NIQE and visual perception are improved.