The Design And Implementation Of Multimode Fiber Spectrometer Based On Deep Learning

Spectrometer is an important spectroscopic equipment in optical instruments,and it is widely used in production practice.Spectrometers have the advantages of high accuracy,wide measurement range,and fast speed.However,traditional spectrometers are large in size,power consumption,and high in cost,which limit the application scenarios and development of spectrometers.So spectrometers are moving towards small size,low power consumption,and low cost.Therefore,it is of great significance to develop a new type of miniature spectrometer.This paper presents a design scheme for a spectrometer that uses multi-mode optical fiber to transmit optical waves to achieve the effect of spectroscopic.Since multi-mode optical fiber carries optical information transmission in different modes and forms a scattering speckle map at the fiber transmission terminal,this paper combines deep learning methods to classify and process the collected optical scattering speckles to achieve the spectroscopic effect.First,use the transmission characteristics of the optical fiber to give a design plan and build an experimental device.Different wavelengths of single-frequency light in the calibrated spectral range are generated using a laser generator,and after transmission through a multimode fiber,the light speckles datasets are collected at the end of the fiber using an infrared camera,and preprocess the datasets.Secondly,introduce the basic structure of the convolutional neural network model and the classic network model,while 2,3 and 4 layer structured convolutional neural networks are built and trained to classify the collected optical scattering speckles datasets.The usability of the multimode fiber optic spectrometer system is verified by experiments,and demonstrated accurate classification effects at different resolutions,matching the most suitable convolutional neural network model for scattering speckle datasets of different resolutions.Finally,considering the practical application of fiber optic spectrometer in biomedicine and other fields,the collected light speckles will be incomplete due to factors such as occlusion when collecting the light speckles.This paper simulates this situation by cropping the datasets with random size and position,and conducted multiple training experiments on the processed datasets,and found the critical point where the residual speckles can still be accurately identified,thus verifying the reliability of the fiber optic spectrometer in practical applications.