Research On Detection Method And Application Of Missing Pin In Aerial Photography Based On Deep Convolutional Neural Network

Due to the low efficiency of traditional manual identification of power inspection image,deep learning is widely used in power inspection,which greatly improves the inspection efficiency of power grid.At present,most of the researches focus on the detection of large targets such as pressure equalizing ring,hanging plate and insulator,and few researches on the detection of small targets at the pin level.The pin plays a very important role in the connection and fixation of power components,so the detection of pin missing is of great significance in maintaining the safe operation of power system.However,the current detection algorithm is difficult to meet the application requirements due to the complex background,small target and obscure characteristics of pins in aerial transmission line images.Therefore,in this paper,aerial pin images are taken as the research object,and deep convolutional neural network is used to detect pin target missing,aiming at how to improve detection accuracy,detection and speed,and applied in transmission line power inspection image detection and management system.The main work is summarized as follows:Firstly,in order to improve the accuracy of pin missing detection,a single-stage target detection method based on attention mechanism was proposed for power inspection images under complex background.In this algorithm,k-mean ++clustering algorithm was used to redefine Anchor,and appropriate Focal Loss parameters were selected.Dense Net is used instead of Res Net as feature extraction network to improve the quality of feature information.Bottom-up connection pathway and channel attention mechanism are introduced in Retina Net structure to improve model generalization ability and high-level structure information.Experimental results show that the proposed algorithm is effective and effective for pin missing detection in transmission lines by constructing pin data set and open data set RSOD.Then,in order to further improve the accuracy and speed of pin missing detection and better reflect the value of engineering application,an aerial image pin target detection method based on cascaded convolutional neural network was proposed.The region of interest is obtained by using small scale shallow convolutional neural network.The deeper convolutional neural network is used to classify and locate the target achieved by the obtained region of interest.On this basis,the nonlinear multilayer perceptron is introduced and the convolution kernel is decomposed,the multi-scale feature graph is fused,and the Angle variable is added into the cross entropy loss function of classification to enhance the ability of small target extraction and improve the detection speed.In the training stage,multi-task learning and offline difficulty sample mining strategies are used to improve the detection accuracy of difficult targets.The experimental results show that the algorithm improves the detection accuracy and speed,and further solves the engineering application problem of pin defect identification in transmission lines.Finally,a set of power inspection image detection and management system is designed and developed.The automatic detection of pins can be realized by using this detection system,which solves the specific problems existing in electric power inspection and embodies the higher engineering application value.