Research On Image Recognition For Typical Objects Of Transmission Line

Making unmanned aerial vehicles equipped with a camera or vidicon, fly along the transmission linecorridor, shoot the lines and its critical components closely, then diagnosis of high voltage transmissionlines automatically by processing and analysis of aerial images using image processing methods, thistechnology is of great significance for achieving the inspection of transmission lines automatically. Inwhich, the Target recognition and defect detection are the key technology and technical difficulties.This paper focuses on the power phase signs recognition, the extraction of the power line, the glassinsulators identification and the defect detection of glass insulator in transmission line images, anddesigned the automated recognition software at last.To identify the power phase signs, it segment the three-channel image in the HSI space at first, then themorphological method is used to denoise or expand the edge, at last, take the intersection of three imagesto got the foreground image, this algorithm can identify the power phase signs and filter out backgroundnoise effectively.To extract the power lines, first, the method utilizes Ratio edge detector to get edge points of lines andHough transform to extract lines; then, the feature detection algorithm is used to extract the power lines.To identify the glass insulators, firstly, the Otsu method was applied to H and S component in HSI spacefor image segmentation; and then, the contour processing algorithm is used directly to got the overalloutline of insulators which obscure each other. In another case, where there is no overlap, the distributionof the contours’ angle and area was calculated by using the histogram method to obtain the position ofsingle glass insulator accurately. In the latter case, the feature detection algorithm, which is based on thespatial sequence relationship of insulators, is designed and used to detect the glass insulators and mark outthe defects.All the ideas in this paper are effective, which is proved by a lot of experiments. These algorithms areused in the automated recognition software, which can satisfy the engineering demand.