Research On The System Of Preventing Outburst Of Dense Transmission Channel Based On Edge Computing

The safe and stable operation of dense transmission channel equipment is related to the stability and security of electric power in the transport link,and is the key to the normal operation of power grid system.At present,the transmission line security methods mainly include manual line patrol and abnormal video surveillance system based on visible light camera.The manual line patrol method has high cost and low efficiency due to physiological limitations.However,the single mode video surveillance is greatly affected by illumination,cannot provide high-quality images in bad environment,and needs to transmit a large amount of video data to the cloud computing platform through the network.Due to the influence of network bandwidth,it is easy to have deficiencies such as delayed warning of external breaking and inaccurate detection of abnormal targets.Therefore,in view of the problems existing in the current intelligent security system monitoring of transmission lines,such as poor environmental adaptability,untimely processing and insufficient accuracy,a study was carried out on the real-time monitoring of the perimeter area of dense transmission channels by combining edge computing and dual-mode monitoring.To solve the problem of difficult identification of suspicious targets caused by complex environment of dense transmission channels and delayed early warning caused by data delay.Firstly,this paper introduces the grid design scheme of the intensive transmission channel anti-breach system in detail,and selects the hardware equipment suitable for the intensive transmission channel through the demand analysis,and determines the software framework and design scheme of the system.Secondly,in view of the low image quality of single-mode image caused by extreme environment in the imaging process,the multi-scale dense connection network image fusion algorithm is studied.Multi-scale convolution was introduced in the coding network to ensure the full use of image features,and Coordinate attention(CA)module was combined to further increase the model fusion effect,so that the fused image reached the optimal in structural similarity,information entropy and difference correlation coefficient compared with the compared algorithm.Finally,the current target detection framework has the problems of huge network volume,difficult deployment on the edge computing platform,and insufficient real-time performance.A lightweight target detection method based on no anchor frame is proposed.By simplifying the complex structure in the model,a high-precision lightweight feature extraction network is constructed,and the label allocation strategy and loss function are improved while the PAN(bidirectional fusion)network is added.The frame rate of the improved network is 25.73 frames /s when the number of parameters is only 2.08 M.The real-time detection of edge devices with limited computing and storage capabilities is realized.