基于Radon和Fourier-Mellin变换的电缆终端红外图像识别  

Identification of Infrared Images of Cable Terminal Based on Radon Transform and Fourier-Mellin Transform

作　　者：牛海清[1] 吴炬卓 许佳[1] 郑文坚[1] NIU Hai-qing WU Ju-zhuo XU Jia ZHENG Wen-jian(School of Electric Power, South China University of Technology, Guangzhou 510640, Guangdong, China Zhuhai Power Supply Bureau, Zhuhai 519000, Guangdong, China)

机构地区：[1]华南理工大学电力学院,广东广州510640 [2]珠海供电局,广东珠海519000

出　　处：《华南理工大学学报（自然科学版）》2016年第8期47-52,59,共7页Journal of South China University of Technology(Natural Science Edition)

基　　金：国家高技术研究发展计划(863计划)项目(2015AA050201)~~

摘　　要：在运行过程中,电缆瓷套式终端的线夹、应力锥和尾管部位可能存在异常发热现象.为了对各部位异常发热的红外图像进行有效识别,文中引入Radon与Fourier-Mellin变换对红外图像进行特征提取.该方法首先对原图像进行Radon变换,再进行解析Fourier-Mellin变换;然后定义与原图像的旋转及与尺度变换无关的不变函数,基于不变函数提取Radon与Fourier-Mellin变换后图像的4种特征.将提取特征向量输入到BP神经网络进行图像识别,结果表明,基于Radon与Fourier-Mellin变换的几何变换不变特征提取方法能够有效反映红外图像特征,具有良好的识别效果,且该方法对噪声具有较强的鲁棒性.Abnormal heating may exist at the clamp, stress cone and tail of a bushing-type cable terminal in opera- tion. In order to effectively identify the infrared images about the abnormal heating at each part, both the Radon transform and the Fourier-Mellin transform are introduced to conduct a feature extraction. In the proposed method, the original image is processed first through the Radon transform and then through the Fourier-Mellin transform. Moreover, an invariant function is defined, which is irrelevant to the rotation of original images and the variation of scale, and four feature quantities are extracted by using the invariant function. Finally, the extracted feature vectors are input into the BP neural network to perform the image recognition. The results show that the proposed method can effectively reflect the characteristics of infrared images, and it has a better recognition performance with a strong robustness to noise.

关 键 词：瓷套式终端 红外图像 RADON变换 FOURIER-MELLIN变换 BP神经网络 特征提取 模式识别 

分 类 号：TM726[电气工程—电力系统及自动化]