Research Of Unmanned Aerial Vehicle(UAV-RRB) Transmission Line Inspection Image Recognition Based On Deep Learning

Transmission line is an important part of power grid,and its running status is related to the safety of power grid.As one of the important tasks of power transmission lines,inspection has gradually developed from traditional manual inspection to UAV inspection,which improves the efficiency of inspection to a certain extent.However,it is difficult and inefficient to select defective data from a large number of UAV inspection data,which also troubles the power grid staff to a certain extent.Therefore,it is necessary to apply the theory based on deep learning to the inspection application of UAV transmission lines to carry out inspection data defect identification.In order to realize automatic identification of defect images in UAV transmission line inspection,based on deep learning theory,this paper uses object detection technology,and on the basis of the YOLO-V5 algorithm model,two modules of attention mechanism and multiscale processing mechanism are proposed to form an improved model which can be applied to defect identification and classification of UAV transmission line inspection data.It is compared with the typical Twostage algorithm model Cascade R-CNN and the typical Onestage algorithm model Center Net.Firstly,transmission line inspection defect data required for defect recognition of UAV transmission line inspection images were collected and experimental data were obtained through data preprocessing.Labelme data annotation tool will be used to make data sets.Secondly,two detection models will be established and compared based on the improved algorithm of YOLO-V5 and YOLO-V5 for five detection targets including insulator burst,bird nest,foreign body on line,balloon hanging on line and kite hanging on line.Finally,a typical target detection model for transmission line inspection based on Cascade R-CNN and Center Net algorithms will be established,and compared with the improved algorithm of YOLO-V5,optimize better model and apply in operational production,the main conclusions are as follows::(1)Through the YOLO-V5 and its improved algorithm for transmission line defect identification,found that the improved model is better the indicators.The Precision index is2% higher than the original model.Recall is 4.7% higher than the original model.The average F1-score index of the improved model is 4.4% higher than that of the original model.m AP,the most important composite index,was 3.1 percenthigher than the original model..(2)Through the Cascade R-CNN and Center Net defects on transmission line,which can identify shortcomings found model stability.Although the average value of each index based on Cascade R-CNN algorithm model can be maintained at good level,the difference between the highest value and the lowest value is large.The difference of Precision index is 8.2%,the difference of Recall index is 23.9%,and the difference of F1-Score index is 14.2%.P-R curve value index difference 11.3%.Similarly,a similar situation can be found for various indexes based on Center Net algorithm model,among which the difference of Precision index is 7.8%,that of Recall index is 12.8%,that of F1-Score index is 6.4%,and that of P-R curve value index is 9.9%.(3)The secondary indexes of the improved model of YOLO-V5 are better than the Cascade R-CNN model and Center Net model.The Precision indexes of the improved YOLOV5 model are 7% higher than the Cascade R-CNN model algorithm and 2.7% higher than the Center Net model algorithm,and the Recall indexes are 10.7% higher than the Cascade RCNN model and 10.6% higher than the Center Net model,respectively.Better overall..(4)The three indexes of the improved model of YOLO-V5 are better than the Cascade R-CNN model and Center Net model.The F1-score and AP indexes of the 5 types of defect detection were all higher than the Cascade R-CNN model and Center Net model.F1-score was9.2% and 7.1% higher than Cascade R-CNN model and Center Net model,respectively.AP outperforms Cascade R-CNN model and Center Net model by 6.6% and 4.5%,respectively..Based on the above analysis,the improved model of YOLO-V5 has better indicators and better effect than Cascade R-CNN model and Center Net model.The single model of UAV transmission line inspection defect identification can simultaneously process large and small targets,and improve the "fast,accurate and efficient" ability of transmission line defect identification.Therefore,in practical work,researchers can carry out research on defect identification of UAV inspection of transmission lines based on the improved model method of YOLO-V5,further improve transmission line defect identification ability,and promote its application in transmission line inspection and other image recognition applications.