Substation Infrared Remote-view Image Recognition

With the development of modern mass production and scientific and technological progress,the demand for electricity is steadily on the increase, and also have a higher demand for the power equipment reliability, economy and stability.In order to ensure the safety of power system,Our country applied the Infrared Technology in the electrical power system monitor,the primary of this technology is that using the image processing and Pattern Recognition technology for device identification and fault diagnosis.This paper is around the technical problems of the area of substation equipment fault diagnosis. Mainly study on the intellectual Methods of recognition. according to the process of pattern recognition diagram. Firstly, do the Infrared Image pretreatment then proposes two segmentation methods, one is based on the morphological edge detection combine with Ostu method, the other is Watershed segmentation based on morphology method. The experiment results show that all methods can obtain the fully image contours, but the Serviceability of the former method is better than the latter. Secondly, used Hu invariant moments to extract image features. As Hu invariant moments is calculated for each pixel,time-consuming Seriously.In order to resovle this , Introduced the concept of line moments and use it to extract the features,but the line moments also has some shortcoming that does not meet the scale invariant,the maximum of the difference between the before and and after scaling change is 12.5934. According to this problem, analysis the Hu invariant moments algorithm and proposes the improved method. The experiment results show that the maximum of the difference between the before and and after scaling change is 0.08622. that proves the improved method is efficiently solve the above problem. lastly, use the KNN classifier to recognition the Substation equipment. the recognition rate is up to 100%. Other, This paper also in-depth study of the human form recognition,and proposes the Humanoid recognition method based on neural network, The recognition rate was 92.857%.