Research On Detection And Location Of Insulators In Aerial Images

As one of the main components of power transmission lines,insulators have a very important influence on the stability and safety of the power system.Therefore,it is of great importance to accurately locate and detect insulators to maintain the security of power system.This paper use depth supervised and unsupervised learning algorithm of aerial image orientation of insulator string for testing,and the algorithm was improved to adjust,in order to improve the detection accuracy.Neural network simulates the cognitive learning process of human brain,and abstractly expresses image and other target data.This paper studies and compares the detection accuracy of supervised and unsupervised deep learning algorithms,and applies them to the insulator targets in aerial photographs to prepare for the fault detection of insulators in transmission lines.Based on convolutional self-encoder,a new training algorithm is proposed in this paper.The error image is processed by using the method of wavelet packet analysis,and then the parameters are adjusted by BP algorithm to reduce the interference of redundancy and noise in the feature information,so as to improve the reconstruction effect of the network.The correlation between kernel functions and the feasibility of the algorithm are analyzed from the perspective of vector space theory.The experimental results show that the wavelet packet decomposition algorithm improves the network performance effectively.However,this method needs to fix the size of the input image,and the definition is not high.Supervised learning uses the latest YOLOv3 deep network model in the field of artificial intelligence to study its principle and accuracy of target detection.A decomposition aggregation algorithm is proposed,which mainly aims at the problems of false detection and missed detection in the process of target detection and location.The real object is decomposed into several continuous and intersecting sub-targets,and the sub-targets are detected and identified.On the premise of guaranteeing the accuracy and speed of sub-target detection,the improved supervised learning network aggregates and redefines the parts belonging to the same entirety by using the characteristic information and meaning of the intersecting regions among the targets,so as to make the detected target area more complete.It has no requirement for the size of input image.At the same time,find out the whole subordinate to the individual sub-goal,and describe its meaning more profoundly.