Dynamic Detection Of Cables In Tunnels Based On Deep Learning

At present,the inspection tasks in the cable tunnel are mostly completed by manual inspection,which not only takes time and labor,but also poses great safety risks.With the development of deep learning and intelligent technology,more and more industrial tasks are completed by robots based on intelligent algorithms.Therefore,this paper designs an orbital robot system equipped with deep learning intelligent algorithms for the difficulties of inspection tasks in the tunnel and the complex environment.The robot inspection strategy is designed,the cable inspection model is built,and the dynamic inspection of the cable in the tunnel is researched.First,the relevant literature is analyzed,the orbital robot is used as the main carrier of the tunnel inspection task,and the relevant physical architecture,software architecture and inspection standards are designed.The robot inspection standards and inspection modes are formulated,the robot inspection strategy diagram is drawn,and the undisturbed switching between inspection modes is improved.An interactive module is designed to ensure good communication between the robot and the main platform.For the main task target cable detection,a special tunnel cable detection model CabNet is designed.Based on the VGG16 network,the full convolution network is used to improve the FCN-8s network,and the hole convolution is introduced to connect with the FCN-8s pooling layer,adding Different sizes of network output layers improve the accuracy of CabNet network detection.The experimental results show that the CabNet model has good cable detection accuracy,which is significantly improved compared to the basic network VGG16 and FCN-8s.Finally,according to the continuity of the cable image in the video,the LSTM model is introduced to extract the temporal characteristics of the cable.Combining the cable position prediction result of the LSTM model and the cable position detection result of the CabNet network,it effectively improves the accuracy of cable dynamic detection.Analyze the impact of different combination models on the accuracy improvement,and use the weighted addition method to form the cable dynamic detection model LS-CabNet.Finally,the common target detection model and target tracking model are compared with the LS-CabNet model.The results show that the detection accuracy of the LS-CabNet network is better than most models,and it has a higher detection efficiency,which is suitable for the cable inspection in the tunnel.