High-voltage Transmission Line Image Based On Deep Convolutional Neural Networks

At present,the conventional power transmission mode adopted in China is high voltage(UHV)long-distance high voltage transmission technology.In order to ensure the safety of high voltage transmission lines,it is necessary to fully consider the operation safety and maintenance of high-voltage transmission lines.The inspection and repair work of high-voltage transmission lines is very dangerous,and the image data processing of the current high voltage transmission lines is still at the level of manual processing,which is difficult to meet the needs of maintenance.In order to ensure the normal operation of the power grid and the personal safety of the maintenance personnel,it is of great significance to use the computer and the drone as the platform to carry the inspection equipment,and to carry out effective artificial intelligence image recognition on the high voltage transmission line.The inspection and repair work of high voltage transmission lines is very dangerous,and the image data processing of the current high voltage transmission lines is still at the level of manual processing,which is difficult to meet the needs of maintenance.In order to ensure the normal operation of the power grid and the personal safety of the maintenance personnel,it is of great significance to use the computer and the drone as the platform to carry the inspection equipment,and to carry out effective artificial intelligence image recognition on the high voltage transmission line.Aiming at the problem of target recognition,an image recognition method of high voltage transmission line based on deep neural network is introduced.Compared with the traditional maintenance method,this method simplifies the data processing of high voltage transmission line patrol,improves the fault identification degree of high voltage transmission line,and guarantees the personal safety of maintenance personnel.It is obviously superior to the traditional maintenance method.Therefore,the identification method of high voltage transmission line based on neural network has high practicability,and has the significance of further research and promotion.Based on the deep convolution neural network learning theory,this paper applies the deep convolution neural network learning to the high voltage transmission line image recognition,in order to improve the recognition performance of specific images.The main work of this paper is:1.This paper elaborates the principle and method of deep convolution neural network Faster-RCNN,the training process of deep convolution neural network learning method,and studies how to select appropriate deep convolution neural network for high voltage transmission line identification.2.In order to solve the problem of long training time of Faster-RCNN network,a Single Shot Multi-Box Detector(SSD)network combined with Batch Normalization batch optimization algorithm is introduced.Because the Batch Normalization batch normalization optimization algorithm can effectively improve the network learning rate,speed up the training,and bring benefits to the flow of information in the network,the dependency dependence of network parameters is weakened.Therefore,this paper optimizes the SSD network and Batch Normalization batch optimization.The algorithm is combined and applied to the specific target recognition problem of high voltage transmission lines.While ensuring the high precision of the detection effect,it effectively solves the problem that the training time of the Faster-RCNN network is longer,and reduces the training time.3.At the same time,this paper also adjusts the network structure of this paper.Because the SSD network will keep the detection windows of all sizes and positions,there are repeated windows and too large windows.Therefore,non-maximal suppression(NMS),retain peripheral window policy and edge approximation strategy are applied to optimize the recognition detection results and improve the recognition accuracy.The network layer,network learning parameters,and other parameters of the network used in this paper are repeatedly debugged,and the parameter range with better effect is obtained.4.Various data enhancement methods for data sets,such as rotation,flipping,random adjustment of HSV color space,automatic scaling,etc.to expand the data set and improve the model’s resistance to over-fitting in training.In the test data set,a good experimental effect was obtained,which further improved the accuracy of the network and obtained a better classification effect model.5.Deep convolutional neural networks are applied to the detection and identification of specific features of high-voltage transmission line images by augmenting the data and adjusting the algorithm framework.The high-voltage transmission line images are respectively labeled with specific targets and candidate regions,and the data set is enhanced by rotation,flipping,random adjustment of the HSV color space,automatic scaling,and the like,thereby obtaining a training set and a test set of images required for the experiment.Experiments show that using the optimized network of this paper can ensure the accuracy reaches a high level and effectively improve the training speed.Compared with the image recognition method based on Faster-RCNN network,the single training time is shortened by 0.39 s,over 33.91%,and the recognition speed is fast.This method can be used in the target recognition system of high voltage transmission lines.6.Applying the content of this paper to the Web,the design and implementation of the specific page of the high-voltage transmission line based on the Web page is designed and implemented.The web terminal mainly includes a data reading module,an image preprocessing module,a neural network module,a training model calling module,and a page design.It realizes the call of the model trained in this paper and makes it easier to operate,and recognizes the target image to get the recognition result.