Research On Spectral Analysis Method Based On Multimode Fiber Speckle

Spectrometers are the basis of spectral measurement,but traditional spectrometers are large and heavy,and cannot be carried around.Therefore,scientists have been studying the miniaturization scheme of spectrometers.Applications in areas such as component testing provide an economical,high-performance tool.At present,there are four main schemes for miniaturization of spectrometers:miniaturization of dispersive components,use of narrow-band filters to replace dispersive components,Fourier transform microspectrometers based on spatiotemporal interference,and speckle calculation-based spectrometers.Among them,the speckle spectrometer based on multimode fiber has comprehensive advantages in terms of resolution,bandwidth,cost,etc.,and is a hot spot in the current academic circles.However,the current research on this type of spectrometer faces difficulties such as low spectral reconstruction speed,low environmental robustness of the spectrometer,and unselectable resolution and bandwidth of the spectrometer.Aiming at the above problems,this thesis has carried out the following two researches:First,through Labview software,an intelligent co-tuning platform between tunable light source,imaging and sampling equipment is built.Based on this platform,a wavelength-speckle mapping fiber structure with multiple bandwidth scales is designed and combined with neural network The speckle spectrometer has an integrated bandwidth and resolution identification,and realizes a speckle spectrometer with adjustable bandwidth in the range of 10nm-160 nm and resolution in the range of0.01nm-0.16 nm.In terms of spectral reconstruction speed improvement and denoising,the influence of noise on reconstruction accuracy is simulated,and a new algorithm for image preprocessing based on convolutional neural network combined with principal component analysis and discrete cosine transform is proposed.Extraction not only filters out high-frequency noise and Gauss-Poisson noise,but also extracts the main information of the image and reduces the data dimension.The algorithm achieves 6times the reconstruction speed improvement and 10 times the noise reduction effect on the simulation data set.Secondly,in terms of realizing the optional resolution of the spectrometer and the enhancement of environmental robustness,the convolutional neural network is used to classify the two-dimensional situation of resolution and temperature,and the specific values of resolution and temperature are obtained in real time according to the measured speckle image.Results By selecting the corresponding calibration matrix,the spectrum can be accurately reconstructed in the temperature range of 20℃-40℃,which effectively avoids the influence of temperature on the accuracy of the reconstructed spectrum,improves the robustness and functionality of the spectrometer,and expands its application scenarios.And the dimension of the convolutional neural network is scalable,which provides an effective method for the next step to overcome situations such as fiber bending and mechanical vibration.