Tunnel Cable Support Identification And Fault Detection

Ensuring the safe operation of the underground power network requires real-time monitoring whether the tunnel cable and related equipment are abnormal.The mainstream tunnel inspection methods are manual inspection and video surveillance systems.However,these two methods currently rely on manual determination of whether there is a fault,so the inspection efficiency is low,and the effect cannot be effectively guaranteed.Based on the robot tunnel inspection,this paper proposes a cable support tilt detection method based on the improved YOLOv3 algorithm,which can automatically detect abnormal conditions on the ground and issue alarms.In this paper,the original YOLOv3 algorithm is improved based on the cable tunnel environment and the characteristics of the large number of brackets,small size,and overlapping perspective.First,in order to make the training loss converge faster,perform K-Means clustering on the labeled ground truth,and use the updated a priori box;then in response to the problems of IoU,GIoU Loss is introduced;the center coordinate loss function and the confidence loss function is Weighted according to the center coordinates of the bounding box to reduce invalid recognition and improve detection accuracy;Finally,the feature extraction network is optimized,the four-fold down-sampling feature map is introduced as the new detection scale and the original 16-fold down-sampling feature map is deleted.Figure.When calculating the inclination of the bracket,the statistical outlier elimination method is excluded,and the interference data is eliminated through K-Means clustering.The algorithm test is divided into two parts,one part is the bracket identification test,which is to experiment to improve the effect of the YOLOv3 algorithm;the other part is the bracket inclination calculation test to verify the effectiveness of the interference data elimination scheme.The final experimental results show that the improved YOLOv3 algorithm achieves the best detection effect when the hyperparameter and is introduced when the loss function is improved.Compared with the original YOLOv3 algorithm,the average accuracy rate of stent recognition at this time is improved by 83.34%To 86.73%,the detection speed increased from 37 frames/sec to 44 frames/sec.At the same time,the interference data removal program has a significant effect,and the accuracy of the bracket tilt recognition has reached 92.83%.