Computer Vision-based Identification And System Development Of Optical Fiber Transmission Box Ports

Now the Internet has entered thousands of households.In order to maintain the stability of the customer's network,telecom operators must regularly test and maintain the optical fiber transmission box.The daily inspection work mainly relies on manually checking the status of each port of the optical fiber transmission box and manually recording the inspection results.Manual detection is not only inefficient,but its accuracy cannot be guaranteed.In response to this problem,thesis proposes to use the current hot computer vision technology to replace manual detection and maintenance tasks.With the continuous development of various computer vision technologies,the use of computer vision to replace manual identifying for the port status of the optical fiber transmission box has become a sure winner.Thesis aims at the problem of long time consumed in the manual detection of the port status of the optical fiber transmission box.Thesis studies a set of computer vision-based optical fiber transmission box port status detection algorithm.The specific design and research work are as follows:(1)Fiber optic transmission box of a mobile company is taken as the research object.The manually collected images may have tilted cabinets,numerous backgrounds,and overexposed or underexposed images,the images are pre-processed.Tilt correction and exposure correction are performed on the image to overcome the interference of human factors on subsequent detection.(2)Taking the pretreated optical fiber transmission box as the research object.a character detection training set are made anda character detector is trained based on YOLOv4.Using the feature that a line of ports of the optical fiber transmission box corresponds to a character,the ports of each line are divided according to the character position.Single row port images will be used for subsequent single port positioning.(3)Take pictures of each row of optical fiber transmission box ports as the research object.A stacked Hourglass Network is improved for treating one-shot localization of each row of port images.Each port image is segmented by ROI for subsequent identification.(4)Take a single port image as the research object.A novel port classification network is designed to classify port images quickly and accurately.Compare it with existing classification networks for recognition effect analysis.The optimal classification network is selected as the port classifier of the system.(5)The state identification algorithm of the fiber transmission box port is tested.Algorithms are transfered to Android system.Choose the most convenient method for algorithm porting.Function of port status detection with mobile APP is implemented.According to the requirements of the enterprise,this thesis successfully realizes the development of the identification algorithm of the port state of the optical fiber transmission box.Test results show that the system can achieve good recognition results.After the algorithm is transplanted to the mobile terminal,it still has a good performance.