Infrared Image Change Detection of Substation Equipment in Power System Using Markov Random Field

An infrared image change detection method based on Markov Random Field (MRF) was proposed to estimate the status of substation equipment in the power system. The method classified changed and unchanged regions between bitemporal images using MRF with k-means clustering initializing the label of all pixels of the sample image. The proposed method used the target pixel and its neighborhood information to realize the determination of the category of the target pixel. In our method, the original bi-temporal infrared images were converted to two gray-level images, and one difference image was obtained by subtracting one gray-level image from another, pixel by pixel. Change areas were then detected on the gray-level difference image using inference techniques on MRF. To demonstrate the excellent performance of our method, comparative experiments were made using the other four classical approaches, including Image Differencing, Image Ratioing, Change vector analysis (CVA) and Principal Component Analysis (PCA). In order to quantify the performance of different algorithms for a quantitative comparison, six performance indexes, i.e. Kappa value, Probability of False detection (PF), Probability of Omission detection (PO), Card Similarity Index (CSI), Classification Error (CE) and Area Error (AE) were adopted in this paper. The experimental results showed that compared with the four classical methods, the proposed method can effectively reduce PO and PF, and improve the overall detection accuracy.