An Improved Framework Of Region Segmentation For Diagnosing Thermal Condition Of Electrical Installation Based On Infrared Image Analysis

• Main Author: Jadin, Mohd Shawal
• Format: Thesis
• Language: English
• Published: 2018
• Subjects: T Technology; TK1-9971 Electrical engineering. Electronics. Nuclear engineering
• Online Access: http://eprints.usm.my/47439/1/An%20Improved%20Framework%20Of%20Region%20Segmentation%20For%20Diagnosing%20Thermal%20Condition%20Of%20Electrical%20Installation%20Based%20On%20Infrared%20Image%20Analysis.pdf; http://eprints.usm.my/47439/

The abnormality of electrical equipment will occur when its internal temperature reached beyond its limits, which can lead to subsequent failure of the equipment. Therefore, early prevention is required in order to avoid this fault while maintaining the reliability of the equipment. This research proposes a new framework of region segmentation and thermal fault detection method for diagnosing the thermal condition of electrical installation by considering both qualitative and quantitative infrared image analysis. Since most of the electrical installations are normally fixed repetitively, a new region detection method is proposed that is able to detect all identical structure of electrical devices within an infrared image. The method employs the combination of the scale invariant feature transform (SIFT) and maximally stable extremal regions (MSER) keypoint detectors for improving the number of keypoint detection. A method for matching and translating clusters is presented by introducing a voting procedure for finding a group of matched clusters. The region detection is achieved by employing a grid approach to divide the repeated cluster before properly segmenting the target region. For evaluating the condition of electrical installation, the effectiveness of thirteen types of input features is investigated. A wrapper model approach is utilized for selecting feature where the multilayer perceptron (MLP) artificial neural network and the support vector machine (SVM) are used to evaluate each of the possible combinations of the feature set. Based on experimental results on the proposed segmentation method, about 94.27 % of the regions were correctly detected with the average area under curve (AUC) value of 0.79 was achieved. Meanwhile, for assessing the thermal condition, it was found that the integration of Tdelta, Tskew, Tkurt, Tσ and dB features yield the best result when classified by SVM using radial basis kernel function. The highest classification rates are achieved at 99.46% and 97.78% of the accuracy and f-score value, respectively.