Research On Detection Method Of Transmission Line Ice Thickness Based On Deep Learning

For the past few years,the rapid growth of our own country’s economy has contributed to the prosperity of the power system.As a result,the construction of power transmission lines in various areas of our country is also in full swing.During the transmission process,various safety accidents caused by icing on the surface of the transmission line have attracted lots of attention.Therefore,timely and effective detection of the power transmission line ice-covered status,ice thickness has paly an significance to keep the safe and stable operation of power transmission lines.Most of the research objects selected are ice-covered simulation using copper wires in the laboratory,and the splitting of ice-covered images is mainly based on image processing.Therefore,two kinds of splitting ways based on the image of ice cover of deep learning power lines are recommended,and the thickness of ice cover is computing in combination with the relevant image processing way.The main work of this paper is as follows.First of all,starting from the basic principles of ice on power lines,the formation process of ice on power lines and the main types of ice on transmission lines are studied.Then,the existing ice thickness detection method and related literature are systematically studied and analyzed,and the image segmentation technology based on deep learning is studied in depth.Finally,two ways based on deep learning are recommended to semantically divide the image of ice cover of power transmission lines and a way for computing the ice thickness cover of power transmission lines.(1)A semantic segmentation method based on improving U-Net.The semantic segmentation ways uses the pre-trained model MobileNetV1 as the encoder of the model,the decoder adds the bulk normalization layer and the drop layer on the basis of the original structure,the network structure is uncomplicated,the number of ginseng is small,the number of network layers is small,the training time is short,the image segmentation result is satisfactory,and the model is hoped to be transplanted to the mobile side.(2)A semantic segmentation method based on DeepLabV3.This way uses Resnet101 as a feature extractor,the network structure is more complex,the network layer is deep,the training time is longer,but the segmentation accuracy is higher and more stable than the above way.(3)A power transmission line ice thickness calculation way.The pixel value and power line info relate to the ice cover image of the ice block of the power line are received,and the ice thickness is computed by the proportional relationship between the pixel value and the real diameter value of the power line.Finally,through the experimental data to verify the semantic segmentation way and ice thickness calculation way proposed in this paper,calculated ice thickness and manual use of cursor caliper measurement value error of less than 1mm,indicating that the detection method in engineering applications has a better practical application prospects,in order to prevent the occurrence of ice cover accidents and conduct the power transmission line ice melting work provides a strong technical support.