Inverse Design Of Mid Infrared Narrow Band Thermal Radition Source Based On Machine Learning

Gas detection in production and life is of great significance to people's life and property safety,industrial development and environmental protection.Many major safety accidents are caused by the failure to detect harmful gases in time and accurately.With the continuous development of science and technology.More efficient gas detection methods are emerging.At present,the common gas detection methods include electrochemical detection,catalytic combustion,gas chromatography and infrared absorption.Among them,the infrared absorption spectroscopy will absorb infrared light of specific wavelength by using gas,which is more efficient than other methods,with high resolution and good selectivity for gas molecules.However,this method is expensive and time-consuming in spectrum analysis.If a narrow band that can emit the same absorption peak as the gas to be detected can be designed,the defect of infrared absorption spectroscopy will be perfectly solved.In order to solve this problem,the forward prediction and reverse design of the mid infrared narrow band thermal radiation are studied in this paper.The traditional methods for the design of forward prediction and reverse structure of photonic devices are finite element modeling(FEM)or FDTD(finite difference time domain implementation)or higher-level topology algorithm.Although these design methods are relatively mature in the field of optics,the optimization cost is still a big problem,especially the reverse design method involves a lot of complex operations,which increases the complexity of the design.Because of its excellent learning ability,neural network can fully realize the prediction and fitting of data,and to a certain extent,it can solve the shortcomings of traditional design methods.In this paper,the mid infrared narrow band thermal radiation light source is selected as the design object.Firstly,the structure design and data simulation are carried out by using the software of Lumerical to obtain the original spectral data of the light source.Then we analyze the spectrum,determine the input and output parameters of the neural network,and finally use Matlab to clean and filter the data to get the available data set.Then we use Python language to construct the corresponding structure of neural network and determine the relevant parameters,and use the processed data set to train,verify and predict the constructed forward and reverse neural network respectively.The neural network is optimized to realize the function of forward prediction and reverse design of mid infrared narrow band thermal radiation source efficiently and accurately.Compared with the existing structure design methods of photonic devices,neural network can realize the design and prediction of the corresponding mid infrared narrow band thermal radiation light source more accurately and efficiently according to our needs,greatly reducing the design time cost.With the continuous improvement and development of neural network,we believe that neural network will be more widely used in the field of materials and optics,with a broader development prospect.