基于FMLNet的光学成像雨(雾)信息自主消除算法

Autonomous Rain(fog) Elimination Algorithm for Optical Imaging Based on Fmlnet

航天发射和回收任务中,通过运载火箭以及无人机获取的光学图像易受雨雾的影响导致成像质量下降。为了同时满足雨纹消除和除雾应用需求,进一步提升雨(雾)图像信息恢复的质量,将特征提取、多尺度映射、局部极值以及非线性回归处理相结合,设计一种新的雨(雾)信息自主消除网络(FMLNet),提出一种基于FMLNet的光学成像雨(雾)信息自主消除算法。使用Maxout单元层生成几乎所有与雨(雾)相关的特征,运用一种新的非线性激活函数(BReLU)以提高恢复无雨(雾)图像的质量,使其特征提取层和非线性回归层与经典CNN网络具有明显的不同。分别对雨(雾)图像数据集进行实验测试,结果表明,算法在峰值信噪比(PSNR)和结构相似指数测度(SSIM)评价指标上均优于其他常用算法。通过各算法处理效果的视觉观察比对,该算法能够很好地进行图像雨纹消除和图像除雾,能将不同雨(雾)场景下的图像恢复至细节丰富的干净场景图,图像复原度更高、视觉效果更好,从而验证了算法的有效性。

In space launch and recovery missions,the optical images obtained by carrier rockets and unmanned aerial vehicles are susceptible to rain and fog,resulting in the degradation of image quality.In order to meet the requirements of rain pattern streaks elimination and fog removal at the same time,and further improve the quality of rain(fog)image information recovery,a new rain(fog)information autonomous elimination network(FMLNet)is designed by combining feature extraction,multi-scale mapping,local extremum and nonlinear regression processing.An autonomous rain(fog)elimination algorithm based on FMLNet is proposed.The Maxout element layer is used to generate almost all rain(fog)-related features,and a new nonlinear activation function(BReLU)is used to improve and to recover the quality of rain-free images.The feature extraction layer and nonlinear regression layer are obviously different from the classic CNN network.The experimental results of rain(fog)image data sets show that the algorithm is superior to other algorithms in the evaluation indexes of peak signal to noise ratio(PSNR)and structure similarity index(SSIM).By comparing the visual observation of the processing effects of each algorithm,the algorithm in this paper can eliminate image rain streaks and make image defogging well,and can restore images under different rain(fog)scenes to clean scenes with rich details,with higher image recovery and better visual effects,thus the effectiveness of the algorithm is verified.