Power transmission line icing detection method based on deep neural network

Abstract

The invention discloses a power transmission line icing detection method based on a deep neural network. Defuzzification, contrast improvement and the like in image processing are fully utilized. Themethod comprises the following steps: carrying out preprocessing operation on field images with uneven quality shot by a camera and taking the images as training and testing data of a deep neural network detection model, thus the model can cope with detection tasks under different field environments, the recognition accuracy of the model is superior to that of an existing power transmission line icing detection method, the detection speed is far higher than that of the prior art, and real-time monitoring of the line condition becomes a reality.

Description

technical field [0001] The invention relates to a method for detecting icing of transmission lines based on a deep neural network, and belongs to the technical field of electric power intelligent operation and maintenance. Background technique [0002] With the rapid development of my country's social economy, the demand for electricity from the industry and the public is increasing year by year. The stable supply of electricity to the society by the power sector has become an indispensable part of our society. To further improve the efficiency of the grid, ensure the safety of workers, Reducing the occurrence of safety accidents is also a new goal of smart grid. Icing on high-voltage transmission lines may cause accidents such as line tripping, disconnection, pole down, conductor galloping, insulator flashover, and communication interruption. Existing methods for detecting icing on lines mainly include manual inspection, ice observation station detection, and traditional im...

AI Technical Summary

Problems solved by technology

Chinese patent document CN106595551B discloses a deep learning-based method for detecting the thickness of ice in transmission line icing images, which belongs to the field of digital image recognition. The accuracy and automation of line ice thickness monitoring, the invention is used for the ice thickness monitoring and overrun alarm of power system transmission lines, including: (1) collecting ice coating images; (2) preprocessing images and establishing data sets ; (3) Establish convolutional neural network; (4) Train and test the model; (5) Extract ice thickness information and send it back to the control center and other five steps

[0007] (3) In a complex transmission line environment, the scales of multiple components of the transmission line in the images captured by surveillance cameras and UAV inspections often change dramatically, so using a single-scale detection model will cause small The problem of invalid detection of large-scale or large-scale targets

[0009] (1) In order to meet the electricity demand of various regions, transmission lines are widely deployed in various natural environments and urban environments, and the actual environments are quite different, so the data sets used for training models often cannot cover all deployment scenarios , and the data on different scenarios that have been covered are prone to uneven distribution

According to our statistics, the problem of data sparsity in different scene categories is serious, and data imbalance can easily lead to low accuracy of deep neural network models in the face of missing or sparse data.

[0010] (2) The brands of cameras deployed in the power grid and UAVs used for inspection are often provided by multiple manufacturers, and the adverse effects of bad weather on the actual transmission line in the environment will make the collection The high-voltage line image data obtained has quality problems such as dark light and jitter blur, which are not conducive to accurately judging the backlog of ice and snow on the line. Therefore, how to use deblurring, contrast enhancement and other technologies to improve image quality as the input of the deep neural network detection model Data is technically challenging

[0011] (3) In the wind and snow environment, the visual distance of the camera is low, and the power transmission lines are often integrated with the surrounding snow environment due to the thinner lines and snow accumulation. It cannot be distinguished, which brings great challenges to the deep neural network detection model

Images