Infrared Image Defect Diagnosis through LAB Space Transformation

Thermal diagnosis based on infrared photos taken by robots is a low-cost measure with promising results in the defect detection of power apparatus. To reduce the work of human investigation on those photos, this paper provides an automatic program by means of image segmentation and template recognition. First, the original RGB (Red, Green, Blue) color of the pixels in thermal images are transformed into the Lab (Luminosity, color vectors a & b) space. Then, we carry out K-means clustering in Lab space, divide the image into K classes, remove the class that represents the background value of the image, and get the image without background. Later, NCC (Normalized Cross Correlation) grayscale matching algorithm is used to find out the location of the device. Finally, the processed images are clustered again by K-means, and the results are divided into different layers, and the variances of L, a, b in Lab are calculated. Fault diagnosis is done by comparing the variance of different levels. The example shows that when the original image is divided into 4 layers, the background can be removed well, and when the number of layers is increased, the corresponding variance of the corresponding variance is significantly reduced to show that the equipment has overheating fault.