针对无人机(UAV)测量中出现的运动模糊问题,尝试将DeblurGANv2网络引入到UAV测量模糊图像的恢复任务中,并设计一种自适应指数移动平均损失函数(Adaptive Exponential Moving Average Loss Function,AEMALF),又将维纳滤波后的图像存在振铃效应进行高频抑制并通过色彩映射等方法恢复原图像部分细节,建立了模拟仿真的UAV测量运动模糊图像数据集.提出了一种改进的对抗网络DeblurGANv2算法和配合维纳滤波预处理的图像去模糊方法.实验结果表明,相较于同类算法,所提算法更能充分挖掘图像多尺度特征,恢复的图像平均峰值信噪比(Peak Signal to Noise Ratio,PSNR)和平均结构相似性(Structural Similarity,SSIM)均有显著提高.
Abstract
To solve the motion blur problem in UAV measurement,the DeblurGANv2 network is introduced into the restoration task of UAV measurement blurred image,and an Adaptive Exponential Moving Average Loss Function(AEMALF)is designed.The image processed by Wiener filtering has ringing effect which should be performed on the high frequency suppression and some details of the original image through color mapping and other methods are restored,and a simulated UAV measurement motion blurred image data set is established.Therefore,an improved adversarial network DeblurGANv2 algorithm and an image deblurring method with Wiener filtering preprocessing are proposed.The experimental results show that compared with similar algorithms,this algorithm can fully exploit the multi-scale features of the image,and the average Peak Signal to Noise Ratio(PSNR)and average Structural Similarity(SSIM)of the restored image are significantly improved.
维普
万方数据