Yolov3 Image Recognition Technology For Substation Personnel Climbing Ladder

The hub of substation as an operating voltage grade change,ensure the safety standards of operation in the substation is very important part of the security and stability of power grid construction,the traditional on-site operation safety regulation,relying on human,however,rely on human to regulate the low efficiency,regulatory manpower cost is high,easy to regulatory blind area problem.In recent years,deep learning has developed rapidly,and more and more target recognition and detection technologies are available.This paper takes the working area in the substation of the power system as the research scene,and carries out in-depth research on the climbing status recognition algorithm of the operators in the substation scene,aiming at the climbing behavior of the operators.Firstly,the traditional target detection method based on HOG is used to identify the operators.Then,the limitations of the traditional target detection method are extended to select the deep learning YOLOv3 with better effect as the detection algorithm for typical power climbing scenes such as substation,and related research is carried out on the basis of it.The main work of this paper is as follows:（1）According to the characteristics of substation ladder climbing data set EPOWER constructed in this paper,the improved K-means++clustering algorithm is used to recalcate the size of the prior box,and a more appropriate size of the prior box is selected to carry out target detection,so that the model frame regression can reach a more accurate position and the model target detection effect can be improved.So as to improve the image recognition algorithm in the power climbing scene applicable level;（2）Improve the feature extraction network Darknet-53 of YOLOv3 model by using DenseNet idea,and replace the RES module in Darknet-53 with dense connection module,so as to make full use of the image feature information and make it more suitable for detecting small and medium-sized targets of EPOWER data set.So as to further improve the image recognition algorithm in the power climbing scene applicable level;（3）A ladder climbing detection algorithm based on YOLOv3 technology is designed.The YOLOv3＿D1 model optimized by K-means++and DenseNet technology is adopted in the algorithm,and the"area cross ratio"IOU₂and the height difference between the ladder climber and the ladder in the ladder climbing behavior are used as the basis for judging whether there is a ladder climbing behavior.The experimental results show that the test accuracy of the algorithm reaches 98.2%,which meets the requirement of substation personnel climbing status recognition performance.