Research On The Power System Thermal Imaging Super-resolution Based On Deep Learning

In the power system,the temperature change of the power equipment is a very important indicator,which is related to the safe and stable operation of the power equipment.Thermal imaging can use infrared photoelectric technology to help humans transcend visual barriers and intuitively see the temperature distribution on the surface of objects,which has more and more needs in the state monitoring of power systems.However,the development materials and technical barriers of high-resolution thermal imagers limit the visualization application of high-resolution thermal imaging in on-line monitoring of power systems.The cheap and effective super-resolution reconstruction algorithm can meet the needs of the power Internet of things while reducing the cost,which has important research value and significance.This paper studies the image reconstruction algorithm based on deep learning,reproduces and improves the current mainstream generative adversarial network,and makes the evaluation index of the reconstructed image better and the recovery of edge texture more perfect.The specific work of this paper is as follows:(1)The edge filtering generative adversarial network(EF-SRGAN)is constructed.Based on the generative adversarial network(SRGAN),the network introduces the idea of soft thresholding,and constructs a threshold learning module in the generator to reduce the heat of power equipment.The influence of noise may be included in the imaging,and an edge loss function is proposed to strengthen the reconstruction of the edge part of the image and improve the texture details of the reconstructed image.The experimental results show that the edge filtering generative adversarial network improves the stability of the network,and the evaluation index of the reconstructed image is improved compared with that before the improvement.At the same time,there is also a significant improvement in the edge data of the reconstructed image,which reflects the recovery ability of the detailed texture and has strong practical value.(2)An edge attention generative adversarial network(EA-GAN)is constructed,which improves the residual network module based on the gradient penalty-based Wasserstein generative adversarial network(WGAN-GP)and adds edge attention The mechanism improves the learning ability of the network,and adds a 1×1 convolutional layer to the upsampling structure,which improves the representation ability of the network.The experimental results show that the edge attention generative adversarial network improves the performance of the network,the peak signal-to-noise ratio and the structural similarity index are significantly improved in the overall image and the restoration of image edge attributes,and the subjective visual effect is better.It has high universality in the field of image processing and has high engineering practical value.