Research On Insulator Recognition Methods In Aerial Images Based On Machine Learning

Insulators easily break down because of being exposed in the wild, so it is essential to monitoring the condition of insulators. With enlarging power system,the way of transmission line inspection by helicopter could take place of manual inspection. Inspectors always make large number of aerial image data and achieve automatic classification and detection with image processing technology. This would improve automatic level, as well as safety and efficiency of power line inspection. This paper mainly researches on insulator recognition and location in aerial photos. It not only provides research basis of later fault detection, but also is important for digitization and intelligence achievement of helicopter power line inspection.Firstly, basic concepts of machine learning and pattern recognition are introduced here, and research status of them are introduced. Secondly, basic procedures of image recognition based on machine learning and related things are talked, including data collection, image preprocess, Haar-like features, moment invariants and evaluation parameters of recognition. Ada Boost algorithm is analyzed selectively. And procedures and error analysis are included. Thirdly, two insulator detection methods, based on Ada Boost classifiers, are proposed in this part.The one is integrated with Bing algorithm of object proposals. The recognition procedures includes object proposals, classifiers recognition and integration of detection windows. This method produce small number of detection windows,compared with sliding windows, achieving fast producing of detection windows.The other one is the method, which combines 3D models and skeleton extraction to realize precise positioning. Recognition procedures includes 3D modeling,classifiers recognition and skeleton extraction. The insulators are located correctly through rough recognition and refine location. As for two methods, the detection process and advanced parts in methods are also introduced with focusing on improvements and experiment performance is analyzed.In summary, this paper studies on insulator recognition in aerial images based on Adaboost algorithm of machine learning and two methods are proposed. Besides,the methods we proposed possess certain stability, which could meet practical needs and has reasonable practical values.