基于改进小波变换与卷积神经网络的干式空心电抗器红外图像去噪方法  

Denoising Method for Infrared Images of Dry-type Hollow Reactors Based on Improved Wavelet Transform and Convolutional Neural Network

作　　者：殷军 殷学功[1] 闫立东 崔岩[1] 张尧 王小朋[1] 李宇航 Yin Jun;Yin Xuegong;Yan Lidong;Cui Yan;Zhang Yao;Wang Xiaopeng;Li Yuhang(State Grid Tianjin Electric Power Co.,Ltd.,Tianjin 300143,China;Hebei Key Laboratory of Power Transmission Equipment Security Defense(North China Electric Power University),Baoding Hebei 071003,China)

机构地区：[1]国网天津市电力公司高压分公司,天津300143 [2]河北省输变电设备安全防御重点实验室(华北电力大学),河北保定071003

出　　处：《电气自动化》2024年第4期90-92,95,共4页Electrical Automation

摘　　要：针对传统小波变换法去除干式空心电抗器红外图像中夹带的噪声效果不理想的问题,提出了基于改进小波变换与卷积神经网络的干式空心电抗器红外图像去噪方法。首先利用卷积神经网络中的残差学习对图像中混合特征信息进行提取;然后通过改进小波变换对图像进行小波分解,并将分解后的分量输入至网络中进行训练;进而通过残差学习增强图像纹理细节信息,解决了传统图像去噪方法的不足;最后进行仿真比较。结果表明,所提方法可以降低网络计算难度,加快训练速度,同时具有良好的去噪性能,优于传统图像去噪方法。A denoising method for infrared images of dry hollow reactors based on improved wavelet transform and convolutional neural network(CNN)was proposed to address the issue of unsatisfactory noise removal in traditional wavelet transform methods.Firstly,residual learning in convolutional neural networks was used to extract mixed feature information from images;then,the image was decomposed by improving the wavelet transform,and the decomposed components were input into the network for training;furthermore,residual learning was used to enhance the texture details of images,solving the shortcomings of traditional image denoising methods;finally,a simulation comparison was conducted.The results show that the proposed method can reduce the difficulty of network computation,accelerate training speed with good denoising performance,which is superior to traditional image denoising methods.

关 键 词：干式空心电抗器 红外图像去噪 改进小波变换 阈值函数 卷积神经网络 

分 类 号：TM93[电气工程—电力电子与电力传动]