Research On Two-dimensinal Denosing Method Of Distributed Optical Fiber Sensor Vibration Signal

In recent years,with the development of the optical fiber communication,the technology of using optical fiber signals for fully distributed monitoring has emerged.Fiber optic sensing determines the state of the object by measuring optical characteristics(light intensity,frequency,phase,etc.)of the scattered light in the optical fiber,and the sensing systems based on various optical fiber loops have big differences in measurement parameters,sensing range and sensitivity,etc.At present,fiber optic sensing based on Phase Optical Time-Domain Reflectometer(?-OTDR)can measure weak vibration signals at multiple points or infinite degrees of freedom,which has the advantages of light weight,small volume,corrosion resistance,anti-electromagnetic interference,high sensitivity when compared with the traditional monitoring methods.The fiber optic sensing is widely used in civil engineering,perimeter security,power equipment,geological survey and other fields.The optical fiber sensing technology based on ?-OTDR senses the physical quantity by demodulating the time-domain phase of the scattered light in the optical fiber while the variations of the phase difference at both ends of the vibration position can be calculated.But in practical application process,due to the characteristics of the optical fiber sensing system and the influence of external monitoring environment,the quality of the collected sensing light signal is poor,which causes the demodulated ?-OTDR vibration signal containing a lot of noise,and it is difficult to accurately reflect the monitored state of the object.The noise that affects the quality of optical fiber sensing vibration signals can be summarized in the following three aspects.(1)Natural noiseDue to the high sensitivity of the optical fiber sensing monitoring technology based on ?-OTDR,the vibration signal generated by any disturbance source can be sensed by the optical fiber sensor,which leads to the actual collected sensing signal containing a large amount of environmental vibration interference,and affects the quality of the optical fiber sensing vibration signal seriously.(2)Attenuate noiseSince there is a certain distance between the monitoring optical fiber and the monitored object,the vibration signal is attenuated severely when it is transmitted to the sensing optical fiber through the medium,which results in weak energy,low amplitude and poor signal quality of the sensing signal collected actually.(3)Modal noiseDue to the difference of coupling degree,coupling area and coupling mode between various sensing fibers,the frequency of optical signals in different fibers drifts,which brings about the formation of parasitic amplitude modulation at the fiber joints,and becomes noise that affects the quality of the optical fiber sensing signal.Based on the above considerations,this article focuses on the optical fiber sensing vibration signal denoising technology.On the one hand,since the collected fiber sensing vibration signal contains time,distance and phase information at the same time,which is consistent with the length,width and pixel value information contained in the gray image,the fiber sensing vibration signal can be processed as an "image";On the other hand,compared with single variable time series processing method,two-dimensional signal processing method can make full use of the potential state and co-occurrence relationship between the data.Therefore,in this paper the convolutional neural network in two-dimensional processing method is used to handle the fiber sensor vibration signal,which is trained by applying the optical fiber sensor signal data set.In the end,the quality of the fiber sensor signal is improved by suppressing the noise in the signal and amplifying the vibration signal.The main content of this paper is as follows:1.Construction of optical fiber sensing vibration signal data setAiming at the lack of optical fiber sensing vibration signal data set,an optical fiber sensing signal data set for network training is constructed.Firstly,the types and characteristics of noise in the optical fiber sensing signal are analyzed which provides support for the establishment of the optical fiber sensing vibration signal data set.Secondly,through comparing the advantages and disadvantages of different normalized preprocessing methods,the mean zero centralization method is used to preprocess the optical fiber sensing signal,which solves the problem that the model is difficult to train because of the small amplitude and inconsistent scale of the optical fiber sensing signal.Finally,based on the o Optical fiber sensing vibration signal obtained in the anechoic chamber,a data set of optical fiber sensing vibration signal o constructed by adding simulated noise.2.Research on denoising method of optical fiber sensor vibration signal based on two-dimensional morphologySince the optical fiber sensor vibration signal is affected by natural noise,which leads to poor signal quality,a morphological-based convolutional neural network signal denoising method is proposed.First of all,through improving the expansion and corrosion operations in traditional morphological filtering,the morphological filtering method has a better noise suppression effect.Secondly,a morphological convolutional neural filter network with an expansion layer and a corrosion layer is construed by defining the minimum pooling layer,and the stochastic gradient descent training method is used to learn the characteristics of noise in the optical fiber sensor signal.Finally,the experiments on the collected optical fiber sensing vibration signals is conducted,and the results show that the method proposed in this paper can effectively filter out the natural noise in the optical fiber sensing vibration signals and improve the signal quality.3.Research on denoising method of optical fiber sensor vibration signal based on enhanced networkSince the optical fiber sensor vibration signal is affected by attenuate noise,which leads to low signal amplitude and weak energy,a signal enhancement method based on neural network with trapezoidal convolution kernel is proposed to enhance the signal amplitude.Firstly,by means of redefining convolution kernels with different sizes,the neural network has different levels of receptive fields;Secondly,a convolutional neural network based on strong supervised learning is constructed using the convolution kernel defined above,which has the up-sampling and down-sampling network structure.The down-sampling network extracts signal features,while the up-sampling network generates enhanced signals.Finally,the experiments on the collected optical fiber sensing vibration signals is conducted,and the results show that the method proposed in this paper can effectively enhance the amplitude of optical fiber sensing vibration signals,reduce the influence of attenuation noise and improve the signal quality.4.Research on denoising method of optical fiber sensor vibration signal based on global threshold segmentationSince the optical fiber sensor vibration signal contains modal noise,which affecs the signal,a noise filtering method of optical fiber sensing vibration signal based on threshold segmentation is proposed.Firstly,a new weight calculation function is constructed,which calculates the probability that each coordinate point belongs to the same area.Secondly,based on the constructed weight function,a new cost function is proposed.and the artificial bee colony algorithm is used to search for the minimum value of the cost function to obtain the optimal segmentation threshold of the optical fiber sensor signal.Finally,the experiments on the collected optical fiber sensing vibration signals is conducted,and the results show that the global dynamic threshold segmentation algorithm proposed in this paper can effectively filter out the modal noise in the optical fiber sensing vibrations signals and improve the signal quality.Based on the above research content,the innovations of this article are summarized as follows:1.A convolutional neural network filtering method based on image morphology is proposed(Mo-CNN).By redefining the expansion and erosion operations in morphological filtering,the "following effect" in morphological filtering is solved,which has a more excellent edge retention effect;At the same time,in order to expand the learning ability of the filter,a convolutional neural filter network based on improved image morphological is constructed.By defining minimum pooling,the network has the same convolutional expansion layer and convolutional erosion layer as traditional morphological filtering.Through training the constructed convolutional neural filter network,the goal of filtering out the natural noise in the optical fiber sensing vibration signals is finally achieved.2.A convolution neural network signal enhancement method with trapezoidal kernel structure is proposed(T-CNN).By sending optical fiber sensor signals into different convolution kernel structures,the characteristic information under different receptive fields is obtained.At the same time,in order to realize the mapping from "graph" to "graph",the constructed convolutional neural network has up-sampling and down-sampling networks.The down-sampling network obtains the characteristics of the signal by learning a specific association mapping relationship,while the up-sampling obtains the amplitude-enhanced signal from the feature map directly using the deconvolution network,so as to achieve the purpose of reducing the attenuate noise in the optical fiber sensing vibration signal.3.A signal filtering method based on global threshold segmentation is proposed.By constructing a weight function to calculate the probability that different coordinate points belong to the same area,which overcomes the problem of isolated points in segmentation,as this function has a faster gradient descent rates.What's more,a cost function is reconstructed for the global segmentation of the signal based on the above weight function,which has obvious extreme points.Through searching the minimum value of the cost function,the optimal segmentation of the optical fiber sensor signal is obtained.Finally,the purpose of filtering the mode noise in the optical fiber sensing vibration signal is achieved.According to the research content and innovation points above,this article is divided into six chapters.The first chapter is the introduction,which explains the background and significance of this research,introduces the characteristics of optical fiber sensing vibration signals,investigates the research status both at home and abroad,and analyzes the advantages and disadvantages of current digital signal denoising technology and convolution denoising technology.Finally,the research content and innovation of this paper is summarized;The second chapter analyzes the characteristics of noise in the optical fiber sensing vibration signal,compares the advantages and disadvantages of different signal normalization methods,and constructs an optical fiber sensing vibration signal training data set,which provides a data basis for subsequent signal filtering and enhancement network;The third chapter proposes a morphological signal filtering method based on the dynamic selection of elements in order to filter the natural noise in the optical fiber sensing vibration signal.Based on the morphological filtering model,a convolutional neural network is constructed and trained.Finally,the experimental results are analyzed and summarized;The fourth chapter proposes a signal amplitude enhancement method based on convolution neural network with respect to the attenuate noise in the optical fiber sensing vibration signal.The proposed network has a trapezoidal convolution kernel which contains an up-sampling network and a down-sampling network.Through training the constructed network,the enhanced fiber signal is obtained.Finally,the experimental results are analyzed and summarized;The fifth chapter proposes a signal filtering method based on global threshold segmentation for the modal noise in the optical fiber sensing vibration signal.The optimal segmentation threshold of optical fiber sensing vibration signal is obtained by searching for extreme value of the newly constructed cost function.Finally,the experimental results are analyzed and summarized;The sixth chapter summarizes the full text and puts forward the direction for further research.