Font Size: a A A

Based On Visible - Near-infrared Spectroscopy And Hyperspectral Imaging For Non-destructive Testing Methods

Posted on:2015-02-15Degree:MasterType:Thesis
Country:ChinaCandidate:J J YinFull Text:PDF
GTID:2261330428477746Subject:Control theory and control engineering
Abstract/Summary:PDF Full Text Request
With the development of society, the research of fast, accurate, andefficient non-destructive testing techniques for promoting sustainabledevelopment in various sectors has very important significance. In recent years,the pattern recognition method combines Near-Infrared spectroscopy andhyperspectral imaging technology are widely used to quickly detect the goodsand obtain the quality information without damaging the goods. Becausenear-infrared spectral data and hyperspectral data are massive high-dimensionaldata, so it becomes very important to reduce these high-dimensional datadimensionality and make the information contained in the data after reducingthe dimensionality does not reduce. To establish the correct model is also veryimportant. In this paper, we combine pattern recognition method andnear-infrared spectroscopy and hyperspectral imaging technology to do the following studies:1. We propose to use autoencoder network manifold learning(AN) toreduce the dimensionality of Visual-Near infrared spectra data of salmonnonlinearly, then use linear discriminant analysis (LDA), least squares supportvector machine (LS-SVM) to establish classification models to classify thequality of the salmon flesh. The accuracy of the classification using the datareduced by principal component analysis (PCA) is lower than that of usingautoencoder network manifold learning(AN).2. We use principal component analysis (PCA) to reduce the dimensions ofhyperspectral image data of transmission fluid, and use the sparse representation(SR) to establish a classification method to classify the transmission fluidvarieties. We also use linear discriminant analysis (LDA) and least squaressupport vector machine (LS-SVM) to establish classification models. Theclassification accuracy of linear discriminant analysis (LDA) and least squaressupport vector machine (LS-SVM) are lower than that of sparse representation.By near-infrared spectroscopy and hyperspectral imaging techniques, wepropose new methods for rapid non-destructive testing of salmon meat and transmission fluid, which can also be applied to non-destructive testing of othersubstances.
Keywords/Search Tags:Visual-Near Infrared Spectra, Hyperspectral imaging technology, salmon, Transmission fluid, Autoencoder network manifold learning, Sparserepresentation
PDF Full Text Request
Related items