Research On Information Exraction Techinique For FTIR Microscopic Image

Fourier transform infrared(FTIR)microscopic imaging technique,with high spatial and spectral resolution,is a new and potential microanalysis technique and has been rapidly developed in recent years.It is nondestructive approach and can provide chemical composition and distribution maps about the surface of a sample simultaneously.Therefore,FTIR microscopic imaging technology has been widely applied in many fields,such as biomedicine,forensic science,food safety and materials science.No matter in what field it is applied,the composition and distribution map of interested object area are needed for further research.However,most of information extraction methods at present are univariate analysis,which sacrifice plenty useful information of the original massive data and constrain the applications of FTIR microscopic imaging technique.Base on the studies of the FTIR microscopic imaging theory and the characteristic of data,the thesis mainly researches the general methods of information extraction.The main achievement is summarized as follows:1.Spectra separation method is used to extract spectra of single component from complex mixtures which containing severely overlapping functions of components.Second derivative for spectra separation is sensitive to noise,so a method of information extraction base on principal component analysis(PCA)-second derivative is presented to solve the problem.Reconstructing spectral data with few principal components associated with chemical compound,the SNR of spectra can be improved to some extent.The information of interested object can be better extracted with robustness to noise by using PCA before second derivative.2.Spectral unmixing is visualizing method,which can offer the microscopic structure and chemical changes of sample.A 2-D image is usually transformed into a 1-D long vector in PCA method before spectral unmixing which loses information of neighborhood pixels and increases the computational complexity.For that reason,a novel spectral unmixing method based on two-dimensional principal component analysis(2DPCA)is developed.It constructs generalized covariance matrix directly using spectral matrix correspond to the column of pixels.With the proposed method,the spectral data can be unmixed with lower computational complexity and higher efficiency.3.Since different features have different contribution to cluster,a new method is proposed,which combine weighted two-dimensional principal component analysis with fuzzy C-means(W2DPCA-FCM)for information extraction.2DPCA is used for rapidly extracting the features of spectral data.Before using the FCM algorithm,the features are weighted by eigenvalues to emphasize their contributions to cluster.Experimental results indicate that W2DPCA-FCM is an efficient information extraction method of FTIR microscopic image since it can reduce the computation time and improve the clustering accuracy.4.A novel maximal standard deviation(MCM)based band selection method is proposed for FTIR microscopic image.The method regards standard deviation for band image as the amount of the information in it.The correlation between band and the selected bands is used as a weight factor for standard deviation in the iterative calculation.By selecting the band with the maximal standard deviation,the typical band subsets with large amount of information and low correlation can be selected.The propose method can reduce the dimension of image data and enrich the data for information extraction.5.Segmentation of FTIR microscopic image by feature extraction and clustering analysis methods can be used for efficient information extraction.However,traditional cluster algorithms are sensitive to initial starting conditions and can be trapped into local optimal solutions.To overcome the drawbacks,we develop a new algorithm(ALO-PSO)in this thesis which improves particle swarm optimization with adaptive local optimization.The local search scope of the global optimal solution is enlarged adaptively along with the increasing number of iterations.Experimental results indicate that ALO-PSO algorithm can improve the rate of convergence and avoid trapping into the local optimal solutions.Combined with PCA,it can extract information from FTIR microscopic image accurately and fast.6.Correct estimation of the number of chemical components is crucial for curve resolution.The resolved spectra will deviate from the real spectra when the wrong number of chemical components is used in curve resolution.The resolution method base on pure variable is proposed which estimate the number of chemical components with pure variable method.Experimental results indicate that the proposed method is an efficient curve resolution method of FTIR microscopic image since it estimate the number of chemical components correctly and extract distribution maps and spectra of the pure components.In summary,deep research of information extraction of FTIR microscopic image had been done in this thesis,and new algorithms had been proposed,experiments had proved that the algorithms can gain good results.