Research On Insulator Detection Technology Of Railway Catenary Based On Deep Learning

Recently,with the rapid development of electrification railway in our country,overhead contact line has become an indispensable part of railway power supply system.As one of the core equipment of overhead contact line suspension device,insulator has the function of electrical insulation and mechanical fixation.Affected by fatigue and wear and the surrounding environment,insulators are prone to failure such as damage and flashover,which leads to the decrease of insulation performance and affects the stability of catenary.Therefore,the detection of catenary insulator state is the key to ensure the stable operation of catenary.Object detection algorithm based on deep learning has become a paradigm in the field of catenary insulator detection.Therefore,this topic carries out research on insulator detection of overhead contact line.The main research work is as follows:(1)Construct catenary insulator data set.In view of the small amount and low quality of the existing catline insulator image data,the data of the original insulator image was cleaned,and then Labelimg annotation software was used for annotation,and the data set was divided.Through image spatial geometry transformation,brightness adjustment,noise enhancement and fog and rain addition,data set of catenary insulator with complex environmental characteristics is constructed for training set.(2)A detection algorithm for catenary insulator based on improved YOLOX-S is proposed.Aiming at the problems such as low identification accuracy and poor detection effect in the detection of catenary insulators,CA attention mechanism was first embedded into the backbone network CSP1＿X structure to strengthen accurate positioning of sensitive areas,so that the neural network could adjust attention on a larger area.Then,Focal Loss was used to optimize the confidence loss of boundary frame to solve the problem of unbalanced background classification and positive and negative samples.At the same time,EIo U Loss was used to optimize the regression loss of boundary frame,making the model more sensitive to the real frame and the predicted frame,and improving the accuracy of the model’s detection of insulators.Experimental results show that the m AP value of the improved algorithm on the contact line insulator verification set is up to90.88%,7.47% higher than that of the original YOLOX-S algorithm,and the detection speed can reach 48.86 FPS.Moreover,the performance of the improved algorithm on the test set is better than that of the current mainstream YOLO series algorithm,which verifies the effectiveness of the proposed algorithm.It is shown that the improved algorithm has better detection effect and better performance in small target and complex background environment.(3)Design the catenary insulator detection software based on Py Side6.In view of the problems that the existing target detection algorithm has strong professionalism and poor intuitionism,and the detection and debugging work can only be carried out through the source code operation,the detection software of catenary insulator is designed based on Py Side6 and the improved algorithm.The software includes user login module and insulator detection module,which can realize local and remote detection of catenary insulators and has good reference value for intelligent detection of catenary insulators.