B. Image Segmentation Methods

The segmentation methods can be divided into: threshold segmentation method, region segmentation method, edge detection method etc. One of the most widely used and the most convenient method is threshold segmentation technology.

By setting a certain threshold filtering, we can detect the heating abnormal in the electrical equipment's image.

Simple threshold method is only applicable to the image which background and target region's gray set has large difference, the problem about this method is prone to excessive segmentation of image. In order to improve the quality of image segmentation, complex threshold methods are widely studied. The Fan Songhai proposed Fuzzy Renyi Entropy and Chaos Differential Evolution Algorithm, it was more efficient than Luo (segmentation based on fuzzy entropy) and Otsu method (maximum between-class variance method) in the same conditions.

C. Image feature Extraction

We extract the target images' features which reflect the characteristics of important and primitive. The important characteristic refers to them as similar aggregation and heterogeneous dispersion in the feature space. The primitive characteristic refers to the characteristic of image absolute strong which does not depend on conditions and environment, specifically not change with the obtained infrared image is different, and minimizes the influences, such as equipment, environmental and other factors on the feature extraction, as possible.

General infrared feature extraction8, as shown below:

1. The highest temperature value of the infrared image φ1;

2. The lowest temperature valueofthe infrared image φ2;

3. The average temperature valueofthe infrared image φ3;

4. Contrast value of the infrared image φ4, which is the ratio of brightness pixel temperature value and the darkest pixel temperature value;

5. The standard deviation of the infrared image φ5, the standard deviation of the target pixel value of temperature;

6. Theratio of brightest pixel number and total number of the target pixel φ6.

D. Intelligent Fault Diagnosis Methods

After processing the image, the next key step is intelligent fault diagnosis. There are many ways to study intelligent fault diagnosis at home and abroad, such as the neural network methods, the fuzzy set theory methods, the application of synthetically artificial intelligence methods, etc.

Due to the neural networks have characteristics of parallel processing, learning and memory, nonlinear mapping and adaptive ability, they are obvious useful in the aspect of electrical equipments fault diagnosis. The neural network consists of self-organizing neural network, feed-forward neural network and support vector machine neural network model. The application of feed-forward neural network is the most extensive, which can be divided into back propagation network (BP network), radial basis function network (RBF network).

Infrared thermal imaging technology has many advantages, which is an important tool for electrical equipment fault monitoring. Intelligent diagnosis is continuous an advancing edge and focus in electric field, to diagnosis fault types precisely is very complex and difficult, but the application of image processing technology and artificial neural network model constantly improving, will make the diagnosis efficiency and accuracy gain significant improve. In this paper, we describe the technology of infrared thermal image processing and neural network model step by step, the advantages and disadvantages are analyzed, the ideas and methods for the further development of intelligent system are provided.