Research On The Surveillance System Of Preventing From External Damege For Transimission Line Based On Deep Learning

Stable and reliable operation of the power system is the fundamental guarantee for the development of the power supply security.In recent years,with the development of the construction of the urban infrastructure,the area of power supply protection is invaded by the various engineering vehicles constantly,especially for bulldozers and excavators,which pose a serious threat to the security of power supply.To prevent the accident of external force damage,the staff has adopted some methods such as manual inspection,installation of infrared sensors and set up laser radar detection devices.However,these methods have disadvantages of high false alarm rate,low recall rate,and vulnerability to environmental influences.In recent years,the State Grid Corporation of China has put forward the strategy of building smart grid and vigorously popularizing the application of video monitoring technology to the operation of the power grid.Under this application background,this paper presents an intelligent monitoring system for power transmission lines against external force damage base on deep learning.The system can realize the functions of real-time monitoring,accurate detection,and timely alarm and so on,which ensures the safe operation of transmission lines.In the proposed intelligent monitoring system,we adopted an object detection model based on deep learning.The model is based on the Faster R-CNN framework,and we modified the network structure of the object detection model.Increase the detection performance of the network by deepening the architecture of the object classification network.According to the concept of transfer learning,we use the fine-tuning method to train the detection model.The pre-trained VGG_CNN_M1024 model is used as the parameter initialization model.We selected a part of the sample images from the images collected in the field,and we annotated the objects in these images.The transmission line scene dataset consists of images and annotation files.The dataset can also be used to train other object detection models which are based on deep learning.We performed a 5-fold cross validation experiment.The experimental results show that the proposed model can accurately locate and identify construction machineries in the transmission line corridor.The average accuracy in89.93% on the test set.