Research On Power Line Foreign Body Recognition Based On Digital Image Processing

The application of unmanned aerial vehicle and patrolling robot has significantly raised the efficiency of power-line-patrolling,however a large number of acquired images needs to be checked manually.A solution to matching the efficiency is to apply computers to recognizing the foreign bodies automatically.Hence,the algorithm for power line foreign body recognition is studied and designed based on digital image processing,which consists of three steps,namely image enhancement,image segmentation and image analysis.The input of image enhancement is a kind of images containing power lines,the results of each algorithm in both stochastic mathematical model and fuzzy mathematical model are compared,leading to the conclusion that fuzzy mathematical model performs better in enhancing the kind of images,on which an enhancement algorithm is based and derived.The input of image segmentation is the output of image enhancement,and the segmentation method based on edge detection is chosen in this step.The application results of various kinds of wildly used spatial edge detection algorithms in enhanced images containing power lines are summarized,which indicates the lack of their applicabilities.The application method of Hilbert transform in 2D signal is therefore studied,and an edge detection algorithm in frequency domain according to phase congruency is further designed.Its better applicability is demonstrated by means of comparison.The processing method of phase congruency is studied,resulting in a better segmentation of edges of power lines and foreign bodies from the backgrounds,as well as the corresponding binary image simultaneously.The input of image analysis is the output binary image of image segmentation.The binary image is further processed on account of region property constraints.Utilizing cognitive physics,the description of data field for binary image is designed,as a result of analyzing principles of image inner features.On the basis of this,a recognition and locating algorithm is excogitated.The performance of operation time and accuracy rate in the whole algorithm process is also experimentally analyzed with sample data.