Research On Automatic Acquisition Algorithm Of Port Resources In Passive Optical Network

Port resource information,such as the number of ports and the occupancy status of passive optical networks,directly reflects the number of fixed broadband network users.This information is an important basis for broadband network operation management and network configuration.Limited by the passive characteristics of port resource information,traditional manual acquisition and management methods have led to the continued low accuracy of passive optical network port resources.With the rapid growth in the number of broadband users,the fast and accurate acquisition of passive optical network port resource information has become an urgent key issue in broadband network operation and management.Therefore,researching the automatic acquisition of passive optical network port resource information and the automatic management of network resources has great theoretical significance and practical application value for fixed broadband operations.Analyses of the characteristics of passive ports,such as planar lightwave circuit(PLC)splitters and optical distribution frames(ODFs),which are mainstream passive optical networks,focus on computer vision-based object detection methods;the main findings of this in-depth study on the automatic acquisition of passive optical network port resource information and automatic network resource management methods are as follows:(1)To solve the inability of the number and occupancy status of passive optical network ports to be automatically recognized using electrical characteristics,the port resource recognition of passive optical networks is transformed into an object detection problem,thus providing a basis for automatically acquiring and managing port resource information.The advantages and disadvantages of different object detection algorithms are compared and analyzed.You Only Look Once Version 3(YOLOv3)is the basic algorithm used to detect the characteristics of passive optical network port information.(2)An improved YOLOv3 deep learning algorithm based on a convolutional neural network is proposed.This algorithm solves the algorithm performance degradation when the PLC port is occluded,the space is small,and small objects are densely arranged at high resolution in practical application scenarios.First,a fourth-scale upsampling feature map is added.A four-scale fusion prediction is formed;this prediction strengthens the ability to extract high-resolution features from an image and enhances the sensitivity of small objects.Second,the PLC splitter dataset is established,and the anchor box dimensions are reclustered by using the fixed port aspect ratio of the splitter to enhance the adaptability of the initial parameters of the anchor box to the specific object of the PLC splitter.Third,the soft nonmaximum suppression algorithm is used to replace the original YOLOv3 nonmaximum suppression algorithm.The improved YOLOv3 effectively improves the detection accuracy of PLC optical splitters,and the accuracy of the improved algorithm is higher than the accuracy of the current mainstream object detection algorithms.The reliability and stability of the new algorithm are evaluated using receiver operating characteristic curve(ROC)and 10-fold cross-validation.(3)Regarding the port expansion of the PLC splitter port in the passive optical network,the ODF port of the uplink passive device has the following characteristics:many ports,a dense arrangement,occlusion,inconsistent models,heterochromatic aging,and greater difficulty in recognition.An improved spatial pyramid pooling(SPP)-based YOLOv3(YOLOv3-spp)deep learning algorithm based on a convolutional neural network is presented.First,the SPP layer is added before the YOLOv3 detection layer to extract and aggregate multiple scales in the feature map.Second,the ODF dataset is established.The k-means++algorithm is used to cluster the anchor box dimensions.Third,the loss function is further optimized to form an improved YOLOv3-spp algorithm.In order to avoid overfitting caused by using fewer samples,an augmentation strategy is designed to expand the dataset,and ODF detection in common scenarios is effectively improved.(4)A cascade model for rerecognizing missing regions is proposed.This model solves the high probability of missed ODF detection and even algorithm failure in difficult scenarios,such as severe occlusion and multiple ports.First,the model designs the processes for port positioning and missed detection according to the ODF size and space;these processes can automatically crop the missed ports locally to generate a dataset of missed ports.Second,based on ResNet-34,a missing area redetection model is constructed,the missing port is captured twice,and the occupancy status is recognized.Third,an end-to-end ODF cascade recognition model is designed.The detection accuracy is improved again by the improved YOLOv3-spp,and the accuracy of the improved algorithm is higher than the accuracy of the current mainstream object detection algorithm.The confusion matrix and F1 score are used to evaluate the classification performance of the cascade model.(5)To solve the low accuracy caused by the manual acquisition of port resources under traditional broadband resource management systems,a design and implementation method for a broadband resource management system based on a passive optical network port automatic acquisition algorithm is proposed.First,a microservice system architecture with high cohesion and low coupling is constructed to enhance the autonomy of the modules.Second,the image recognition module based on the automatic port acquisition reconstructs the network access,opening,change,and exit processes.Third,based on the new services process,the corresponding microservice response cluster is designed to realize dynamic resource management and control.Fourth,the broadband resource management system and the currently used customer relationship management,service opening,installation and maintenance scheduling,and integrated resource function modules complete the standard interface interaction,thereby improving the accuracy of the passive optical network port resources.Through the abovementioned research on the automatic acquisition algorithm of port resources in passive optical networks,the application of computer vision in optical fiber communication has been expanded.Experiments show that the improved YOLOv3 algorithm designed in this dissertation has a detection accuracy of 97.16%for the PLC splitter port,which is 4.15%higher than the original YOLOv3.The end-to-end cascade recognition model based on a convolutional neural network design has an accuracy rate of 95.02%for ODF port detection,which is 7.89%higher than the original YOLOv3.The broadband resource management system based on the port automatic acquisition algorithm realizes the evolution of port resource management from manual to automatic,thereby effectively improving resource accuracy and production efficiency.This system provides a new effective way and technical support for fixed broadband network operators to reduce network investment waste.