Research On Principal Component Feature Extraction And Target Clustering In Infrared Spectroscopy

Infrared spectrμm can reflect the infrared radiation characteristics of the target,but its noncontact measurement method is inevitably affected by the environment and other interference factors,resulting in the misclassification of target recognition and clustering.In this paper,based on the comparative analysis of principal component analysis（PCA）,de-noising self encoder（DAE）and linear discriminant analysis（LDA）feature extraction methods,the infrared spectrμm feature vector is constructed by fusing the features extracted by the three methods,and then the K-means and convolutional neural network are used to cluster the feature vector to distinguish the target type,and finally an infrared spectrμm principal component feature is established Feature extraction and target clustering method framework.The main contents include:Target infrared spectrμm test and calibration.FTIR spectroradiometer is used to measure the infrared radiance of different targets at different temperatures and wavelengths;the interference factors are analyzed,and the calibration method of infrared radiance spectral data is studied.The complete spectral radiance data curve of different targets at different temperatures and wavelengths is obtained,which provides an effective data basis for feature analysis and clustering.Principal component feature vector extraction of infrared spectrμm.Firstly,the feature vector of spectral radiance data is established,which can fully contain the effective information of 2-14 μ M band of infrared radiance spectrμm.Secondly,principal component analysis（PCA）,de-noising self encoder（DAE）and linear discriminant analysis（LDA）are studied to extract the principal component features of spectral feature vector and reduce the feature vector Dimension,reduce the impact of redundancy.Research on target infrared spectrμm clustering method based on feature vector fusion.PCA mainly reflects the low correlation characteristics of feature vectors,DAE has the characteristics of obtaining the original data information as much as possible,LDA can reflect the basic characteristics of different tag targets.By effectively fusing the feature vectors extracted by the three methods,we can not only explore the original nature of the data,but also reduce the relevance of the data,so as to obtain a set of feature vectors that contain more comprehensive infrared spectral information.Combined with K-means method,target clustering based on infrared spectrμm feature vector is realized.Through the simulation of the infrared spectra of the measured blackbody and a new type of steel plate coating,the results show that the K-means clustering method based on feature vector is more accurate.Research on feature vector clustering of target infrared spectrμm based on convolution neural network.In order to avoid the influence of redundant information caused by feature fusion and improve the accuracy of classification,convolution neural network is introduced to study the surface features of data independently,extract and map the features hierarchically,and finally realize clustering.Compared with K-means clustering simulation results,it shows that the multi-layer adaptive feature vector extraction based on convolutional neural network for fusion vector,and then classification,can more effectively retain the independent principal component features in the feature vector,and make the classification more accurate.Based on the content of this paper,the software system of MATLAB app designer is designed to realize the selection of sample object file,the feature extraction of object data and the result display of different clustering methods.This paper designs a computer operation interface which uses graphic operation,displays the results of target feature extraction and clustering analysis,and realizes the research method in this paper,which is more practical.