Optical Fiber Spectrometer Based On Neural Network Algorithm

The spectrometer is an important optical instrument.Modern spectrometers are used in many fields,such as metallurgy,chemical industry,medical treatment,food hygiene and so on.Among them,some special-purpose spectrometers do not need to restore the spectral lines,but only need to classify specific substances or detect the existence of certain substances,which can be achieved by the classification function of neural networks.In this article,the researchers try to combine traditional industrial products such as special-purpose spectrometers with emerging methods such as neural networks.Coherent light will produce module interference in a multimode fiber.Different models have different propagation speeds and different refractive indices,so the phases at the output end are also different.When the phases difference at the output are constant and there are enough modes,a stable and obvious interference pattern will be formed.The interference process in a multimode fiber can be affected by many ways,including the deformation of the multimode fiber,the polarization state of the light,and the different wavelength of the light.All of them could lead the change of the interference speckle pattern.Artificial neural network is a kind of deep learning method,which imitates the neural network model of the human brain,and can simulate many non-linear laws of nature through non-linear activation functions.Neural networks emerged as early as the1980 s.But because of the lack of computing power of the computers at that time,they didn’t get many applications.In recent years,due to the improvement of computer computing power,artificial neural networks have been used in pattern recognition,automatic control and many other fields.The multi-mode fiber spectrometer is manufactured based on the principle that different light produces different interference speckles on the output end face after being transmitted through the multi-mode fiber.According to the principle of the correspondence between different spectra and different interference speckle patterns,the spectral information contained in the interference speckle patterns can theoretically be identified and classified.Based on the corresponding relationship between the two-dimensional distribution of the spectrum and the interference speckle pattern,the spatial intensity distribution of light waves of different wavelengths is used as input,and the corresponding spectral information is used as output.In this way we could achieve spectral-to-spatial mapping and set the neural network like that.So that its algorithm can finally recognize its spectrum and classify different substances that have never been seen before.