特高压换流站保护系统全景监视图像超分辨率重建方法研究

Research on Super-resolution Reconstruction Method of Panorama Monitoring Images of Extra-high Voltage Converter Station Protection System

针对特高压换流站全景监视系统运行环境导致的视频图像抖动、镜头出现积灰等问题,以及基于深度学习的高分辨率图像重建算法存在细节特征失真和计算复杂度较高的缺陷,提出一种基于多尺度卷积块和残差网络的图像超分辨率重建方法,通过增加具有较小内核的深度卷积层来获取图像的鲁棒细节特征,并在训练过程中加入残差网络,加快网络收敛速度,解决消失梯度,改善图像重建质量。对部分标准数据集和特高压换流站全景监视图像数据集进行了图像超分辨率重建和目标识别实验研究,与超分辨率卷积神经网络(super-resolution convolutional neural network,SRCNN)和快速超分辨率卷积神经网络(fast SRCNN,FSRCNN)方法相比,所提算法的结构相似指数均值分别增加了0.0043和0.0298,峰值信噪比分别提高了0.17 db和0.83 dB。实验结果表明所提方法重建了细节信息更逼真的高分辨率图像,可以满足换流站全景监视的需求。

Aiming at the problems of video image jitter and lens dust accumulation caused by operating environment of the extra-high voltage converter station panoramic monitoring system,as well as the defects of detailed feature distortion and high computational complexity of the high-resolution image reconstruction algorithm based on deep learning,an image super-resolution reconstruction method based on multi-scale convolution block and residual network is proposed in this paper.By adding the deep convolutional layer with a smaller kernel,the robust detailed features of the images are obtained.The residual network is added in the training process to accelerate the network convergence speed,the vanishing gradient is solved,and the quality of image reconstruction is improved.Experiments on image super-resolution reconstruction and target recognition are both carried out on Set5,Set14,and Urban100 standard datasets and the extra-high voltage converter station panoramic monitoring image dataset.Compared with RSCNN and FSRCNN,the SSIM mean of this algorithm increases by 0.0043 and 0.0298 respectively,and the PSNR mean increases by 0.17 dB and 0.83 dB respectively.The experimental results show that the high-resolution image with more realistic detail information reconstructed by this algorithm,and panoramic monitoring of converter station are satisfied.