Research On Quantitative Analysis Algorithm Of Primary Elements Of Martian-like Mineral Based On LIBS

Laser-induced Breakdown Spectroscopy(LIBS)has been widely used in many fields due to its advantages of no sample preparation,ease of use,and the ability to detect target components at a distance.In the field of deep space exploration,especially the Mars exploration mission,LIBS technology has a pivotal position.In the past decade,NASA and China have launched the Curiosity.Perseverance,and Tianwen-1 Mars probes,all of which carry LIBS equipment(ChemCam.SuperCam,and MarSCoDe)for the detection of the Martian surface composition.Reducing the influence of the matrix effect and improving the accuracy of LIBS quantitative analysis has become a major challenge in the Mars exploration mission.In addition,the problem of insufficient labeling information and difficulty in quantitative modeling is another major challenge for the quantitative analysis of LIBS,especially for the MarSCoDe data.To solve the above problems,the following parts of work are carried out in this paper:Firstly,aiming at the problem of insufficient accuracy of LIBS quantitative analysis,this paper uses SuperCam calibration data set to conduct quantitative analysis in the form of full supervision.In this paper,the Transformer encoder structure is fused with the convolutional structure to build the SiT(Spectral vision Transformer)quantitative analysis model,and the effects of different activations,different optimizers,different initial learning rates,anddifferent mput methods on the model performance are compared.Then,the STT model is compared with Partial Least Square Regression(PLSR),Support Vector Regression(SVR),and Convolutional Neural Network(CNN)algorithms.The results indicated that the prediction accuracy of SiT model was better than the other three models.Meanwhile,the SDU-LIBS dataset composed of 50 geological standard samples was prepared in this paper,and the average filtering,denoising,discontinuous,and other preprocessing operations were carried out on the dataset,and the SiT quantitative analysis model was successfully established on the dataset.The results show that the prediction accuracy of the SiT model is still better than that of the traditional quantitative analysis model on the small sample data.Second,this paper addresses the problem of insufficient label information of LIBS spectral data,which makes it difficult to model quantitatively,and uses a semi-supervised quantitative analysis model based on the Π-Model architecture for semi-supervised quantification.First,this paper analyzes ChemCam data and MarSCoDe data,and then builds the COREG model,ΠModel model,and Mean Teacher model based on ChemCam data,and finds the best overall performance of the Π-Model through the comparison of the three.Meanwhile,after experimental comparison,the Π-Model outperforms the traditional fully supervised quantitative modeling(Linear SVR).Finally,this paper applies the II-Model model to MarSCoDe data to carry out semi-supervised quantitative analysis tasks successfully,and 8 major elements(Ti,Al,Fe,Na,K,Mg,Ca,etc.)have satisfactory results except for Si elements.Last,the LIBS quantitative regression software was designed by combining various preprocessing methods and quantitative modeling methods.The software integrates pre-processing functions such as spectral data collection,spectral noise reduction smoothing,de-continuity,and some quantitative analysis functions,including quantitative modeling and model optimization.And the quantitative analysis was carried out for SuperCam data to achieve the purpose of software design and validate the software process.