Research On Intelligent Detection Technology Of Optical Fiber End Face Based On Feature Fusion

As the basic component of optical system,the quality of optical fiber determines the performance of the whole transmission system,and the optical fiber end detection technology is an important symbol to measure the quality level of optical fiber.At present,the detection of optical fiber end face still depends on the traditional human eye detection,which has the disadvantages of low efficiency,low precision and long time consuming.With the rapid development of the information age and the continuous improvement of the intelligent level of the factory,the traditional artificial detection technology is about to usher in a major change.In order to save manpower,improve detection efficiency and meet the needs of intelligent development,the intelligent detection method of optical fiber end face is studied,and the intelligent and automatic detection of optical fiber quality can play an inestimable role in the optical fiber production industry.Based on the deep research of object detection,image recognition and defect classification,this paper presents an intelligent detection algorithm of optical fiber end face based on feature fusion technology to realize real-time detection of optical fiber end face.The main research contents and achievements are as follows:(1)According to the characteristics of simple background and different defect size of fiber end face,the lightweight basic network is used to replace Darknet-19 basic network on the basis of YOLOv3 network,and the detection speed of the network is improved to meet the requirement of real-time detection.A larger feature map of the receptive field is added to the base structure of the feature pyramid FPN,and the lower level feature information is fused to improve the detection effect of the small target.(2)In order to solve the problem of information loss caused by the fusion mode of feature pyramid direct splicing,a new adaptive feature fusion method is adopted ASFF,which allows the neural network to learn the weight parameters better to find a better fusion mode,and to fuse the features of different layers together in the best way,which greatly improves the robustness of the fiber end detection model.(3)The optical fiber end detection model is trained by pre-training and fine-tuning.Through continuous parameter tuning,the detection accuracy of optical fiber end defect data is 92.4% m AP,4.2% higher than YOLOv3 algorithm m AP,7.7% higher than Faster RCNN algorithm The detection speed is 41.7 FPS,compared with the 29 FPS of the YOLOv3 algorithm,and the detection speed is much higher than that of the two-stage algorithm.The experimental results show that the fiber end detection model can not only meet the real-time requirements,but also improve the detection effect of small defects,and improve the recall rate and detection accuracy of target detection.(4)To build an optical fiber end detection system by studying its performance requirements.An intelligent detection system is designed to identify and classify five kinds of optical fiber end defects,save image data and defect information to database,and display defect information to man-machine interface.