The Three-dimensional Fluorescence Spectrometry Analysis Methods Based On Support Tensor Machine And Its Application To Mycotoxins Detection

The three-dimensional fluorescence spectroscopy can describe fluorescence intensity that responds with changes in excitation and emission wavelengths, which can completely describe the substance of the fluorescence characteristics of the spectrum and is a valuable fluorescence fingerprint technology. However, the quantitative analysis methods based on three-dimensional fluorescence spectra are mainly the first-order correction method based on the vector and the high-order correction method based on the tensor. With the three-dimensional spectral data matrix is essentially tensor, a new three-dimensional spectral quantitative analysis method, based on support tensor machines (STM), is presented. The presented methods preserve the intrinsic structure of the three-dimensional spectrometry, and reduce the modeling parameter to avoid the over-fit problem in the case of small samples, and do not need to estimate the component number like the high-order correction method based on the tensor. The main work of this thesis is organized as follows:1. Introduce a regression modeling method based on support tensor machines(Support Tensor Regression,STR) and take detailed derivation and description. Then take comparison experiments with PCR, PLS, SVM, PARAFAC, N-PLS to verify the generalization capability of STR. We also do the simulation experiment with the different parameter of C and ε, the experimental results demonstrate that the STR is not sensitive to C and ε2. For the nonlinear three-dimensional fluorescence spectrum, we propose a new kernel function for tensors. It’s a good method to extend the Support Tensor Regression(STR) to the Kernel Support Tensor Regression(KSTR). Then we also take several experiments like the STR to verify the capability of KSTR.3. Compound20mycotoxins experimental samples and get the data of three-dimensional fluorescence spectra. And then the proposed method is applied to detect the mycotoxins with three-dimensional fluorescence spectroscopy, the result confirmed the validity of the study.