Research On Transmission Line Damage Object Detection Based On Deep Learning

The normal operation of transmission lines plays a vital role in ensuring production and life.A series of transmission projects such as West to East power transmission rely on long-distance transmission lines across provinces and cities.Inner Mongolia is an important national energy base,with the first power transmission capacity in China.It is the main province in the northern channel of West to East power transmission,and also the province with the longest total mileage of transmission lines in China.Under the background of " carbon peaking and carbon neutrality goals",the mileage of transmission lines has further increased,which has brought greater challenges to ensuring the normal operation of transmission lines.Under the continuous load power transmission environment in the field,the transmission line is prone to injuries such as nonstandard grounding body,disconnection of grounding body,missing gasket,locking pin,loose nut,loose prestranded wire,and damaged foundation protective cap,which affects the safety of the transmission.At the same time,because the transmission line is located in various complex geographical environments and has a heavy inspection burden,the early transmission line inspection relies on the passive stress maintenance mode of manual visual inspection,sampling inspection,and subjective qualitative inspection,which can not meet the current requirements,and there is an urgent need for an efficient inspection method.The UAV’s flexibility,precision and accuracy,low operation cost,and fast processing of object detection algorithm can provide a new inspection method.In this thesis,UAV is used for transmission line inspection,and the inspection object detection of multiple damage types in the inspection process is studied.Each damage is independently distinguished,finely classified,and quantitatively identified,to promote the intelligent development of transmission line inspection.The main work of this thesis includes:(1)Aiming at the lack of high-quality line damage image data set,the transmission line damage object detection data set is made.Based on 1012 transmission line images taken in Inner Mongolia,a data set of transmission line damage object detection was established after manual annotation,including typical categories such as abnormality of grounding body,corrosion of metal fittings,and damage of basic protection cap and abnormality of nuts.It is superior to the existing data set in image resolution,coverage category,and annotation quality.(2)Proposed an object detection algorithm suitable for patrol personnel in remote areas with limited computing power: Firstly,features are extracted based on the interleaving feature extractor and memory module of Mobile Net V3 to achieve feature fusion.Then,the lightweight network is designed by combining the deeply separable convolution and inverse residual structure.At the same time,the swish function is replaced by the h-swish function for deployment in the mobile terminal.Experimental results show that,in terms of accuracy,the corrosion of the metal is0.916,the abnormality of the grounding body is 0.905,the damage of the basic protection cap is 0.851,the abnormality of the nut is 0.673,and the abnormality of the lock pin is 0.522.The detection speed of 30+FPS is achieved at the mobile terminal.The algorithm can be generalized and compatible with multi-point corrosion of metal fittings and damage of basic protective caps.(3)Object stabilization algorithm is introduced to solve the problem of jitter in the boundary box of damage object detection.The boundary frame wobble around the object center in the object detection of the video shot in inspection,which is not conducive to inspection,and loss determination of the detection results in multiobject scenarios.Improving the stability of object detection enables inspection personnel to more accurately assess the damage situation of transmission lines.