Research On Detection And Fault Identification Of Transmission Line Components Based On Deep Learning

Transmission line inspection plays a vital role in maintaining the normal operation of the power grid,but transmission lines are distributed in various complex geographical environments,and the variety of components and faults to be inspected on the line are huge,which brings great Work load.At present,the scale of China’s power grid is expanding,and the traditional inspection method can no longer meet the requirements.It is urgent to find a new efficient and convenient inspection method.The combination of the cross-geographical characteristics of the UAV and the efficient image processing features of computer vision technology provides a new solution for transmission line inspection.Based on this background and facing the development direction of front-end inspection of unmanned aerial vehicles,this paper conducts research on the positioning and identification of various targets to be inspected in line inspection,which plays an important role in promoting the intelligentization of transmission line inspection.The main work of this paper is as follows:(1)For the shortage of power data sets used for deep learning training,sort out the power component detection data set and insulator self-detonation fault data set.Research related theories of deep learning,verify the performance of various classic target detection algorithms on power component data sets,and choose a benchmark model to lay the foundation for further research.(2)Propose a lightweight feature fusion detection model(LFF-DM)suitable for real-time detection of UAV front-end.Firstly,a priori anchor is set by the improved k-means algorithm,and then the global lightweight network structure is designed by combining deep separable convolution and inverse residuals,and at the same time,the h-swish activation function is introduced for embedded devices.The experimental results show that the detection speed of 25 FPS and the detection accuracy of 90.48% m AP can be achieved on the embedded device NVIDIA Jetson AGX Xavier,which is suitable for real-time and accurate inspection of the transmission line mobile terminal.(3)Analyze the fault characteristics of different components and adopt different fault identification schemes,which provides new ideas for fault identification of transmission lines.For the target to be detected that the fault is not easy to distinguish,the pre-detection step is used to eliminate the background interference,the target area is limited to the specific component area,and then the two methods of detection and classification are used to identify the fault.For the classification scheme,this paper proposes a multi-feature fusion bilinear convolutional neural network(MFFB-CNN)based on fine-grained classification.Through multiple feature fusions in the feature extraction process of the two feature extraction networks,the amount of information contained in the feature vectors is improved,and finally a better classification result is obtained.