Abstract: Infrared image segmentation of power equipment is the basis of pattern recognition for power equipment and infrared fault diagnosis. The Chan-Vese model can effectively segment images with strong noise and blurred edges, but its segmentation speed is slow, and it can not eliminate the unrelated background effectively in the segmentation of power equipment infrared images. In this paper, an improved Chan-Vese model is proposed. A binary function is used as the initial level set function instead of the distance function, and multiple initial contours are applied. Moreover, the data term of the Chan-Vese model is simplified, and the length regularization term is replaced by a Gaussian kernel function. Due to these modifications, a much larger time step can be adopted in the iterative process and the curve evolution speed can be accelerated. Experimental results on power equipment infrared image segmentation prove that the new model has higher segmentation speed and better performance to eliminate the unrelated background compared with the traditional Chan-Vese model.

Key words: power equipment infrared image segmentation, Chan-Vese model, level set evolution, Gaussian

摘要： 电力设备红外图像分割是电力设备模式识别和红外故障诊断的基础。Chan-Vese模型能够有效分割含强噪声和边缘模糊的图像，但其分割速度缓慢，并且在分割电力设备红外图像时不能有效消除无关背景。提出一种改进的Chan-Vese模型，采用多个初始轮廓，并采用二值函数代替距离函数初始化水平集函数；同时对Chan-Vese模型的梯度下降流提出改进，简化其图像数据项，并用一个高斯核函数取代长度正则项。改进的模型不仅方便计算，而且可以在迭代过程中采用更大时间步长，加快曲线演化速度。在对电力设备红外图像的分割实验中，证明了相比Chan-Vese模型，新模型分割速度明显提高，并且具备较好的消除无关背景的性能。

关键词: 电力设备红外图像分割, Chan-Vese模型, 水平集演化, 高斯核函数