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 operation.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.