Research On Defect Identification Method Of Key Components Of Transmission Line Based On Deep Learning

Safety of transmission lines is the basis for securing the stable development of China’s economy and also an important guarantee for the safety of residents’lives.Based on the theory of deep learning,this thesis conducts researches on methods for recognizing defects on key parts of transmission lines and its main research contents and works are as follows.This thesis introduces first the major defect classifications on key parts of transmission lines and then the major components of the convolutional neural network（CNN）,analyzes the object detection technology based on deep learning,prepares the data set of defects on key parts of transmission lines which are pre-processed so as to highlight the object positions of defects on key parts.Then it takes Tensorflow as a platform to conduct two-stage object recognition of defects on key parts of transmission lines according to different regions,puts forward the Faster R-CNN basic framework which extracts the network according to characteristics of defect Inception-ResNet-V2,respectively conducts feature extractions,object positioning and classified operations on images of transmission lines and adjusts the network’s structure and optimizes parameters so as to obtain a defect object recognition model.This optimized algorithm enhances the recognition ability of network models on small defect objects in the images of key parts of transmission lines in this thesis and with comparative analysis of the experimental results,it proves that the method mentioned in this thesis has good recognition effects with a recognition accuracy as high as 98.65%.To improve the recognition speed,this thesis conducts researches on the one-stage object detection technology SSD algorithm based on regression,puts forward the lightweight neural network SSD-MobileNet-V1 model,adjusts the settings of the default box and applies the network in object recognition of defects on transmission lines.Compared with the traditional SSD algorithm,though the recognition accuracy declines slightly,the inspection speed is two times of that of the traditional model which can completely satisfy the real-time demands.At last,it designs the system software for recognizing defects on key parts of transmission lines based on Qt and in the UI interface of Qt,it completes the layout and design of the software interface;it realizes functions including selecting,pre-processing and saving local images;it completes the design and realization of system software for recognizing defects on key parts of transmission lines with the algorithm in this thesis.