The Technology Of Spectral Recognition Based On Statistical Machine Learning

With the launch of artificial satellites,manned spacecrafts and the international space stations and other spacecrafts,spatial target recognition technology is the precondition for further development of space resources.The effective identification of the surface material of the space target has important practical significance and application value to further identify the target.Scattering spectra can effectively characterize the surface characteristics of the samples under test,Statistical machine learning provides a technical means to solve the problem of difficult classification and recognition between samples.Based on scattering spectrum and four statistical machine learning methods,the classification and recognition of spatial object material are studied in this paper.The researches are as follows:1.Building material measurement system,the experiment can detect the scattering spectrum of material under multiple perspectives.The measured scattering spectrum was pretreated,which was denoised,calculated the BRDF and normalized,and the material database was established.2.The algorithm of the Naive Bayesian classifier,the K Nearest Neighbor,the Error Back Propagation Neural Network,the Convolution Neural Network are established,and the programming based on MATLAB is implemented.3.Based on the Naive Bayesian classifier,the K Nearest Neighbor,the Error Back Propagation Neural Network and the Convolution Neural Network,the pre-processed material scattering spectra are classified and identified,and the recognition results were analyzed and compared.The research results show that:(1)When using the K Nearest Neighbor,with the method of embedding Angle Cosine and Euclidean distance,by fully considering the linear features and the amplitude spectrum curve,which has the characteristics of high precision and low time consuming.The method has certain applicability in the field of identification based on the scattering spectrum.(2)Because of its special network structure,Convolution Neural Network has the characteristics of low time consuming and high precision,this method has the advantage and applicability different from other methods in the field of classification and recognition of spatial objects with large data volume.