Research On Methods And Applications Of Feature Extraction And Classification Of Hyperspectral Microscopic Images

The technology combining hyperspectral imaging and microscope used in biomedicine is a medical imaging method developed lately.It can take images which involve simultaneous pictorial and spectral information on near-distanced biological tissues.Since it can provide more plentiful chemical features for biomedical analysis than 2D images,it will be greatly helpful for pathologists to do the research on tissues and cell slices from a new perspective.Based on the hyperspectral microscope system invented by ourselves,the goal of this thesis is to specialize the research on feature extraction and blood cell classification of hyperspectral images of biomedical samples by transmission imaging.The main achievements are summarized as follows:1.Feature extraction of hyperspectral images was carried out using non-dominated sorting genetic algorithms(NSGA).An adaptive NSGA algorithm called ANSGA was proposed here to solve the problem.Based on the statistical analysis of hyperspectral images of diabetic retinopathy and leukemia blood cell,multi-objectiv genetic algorithms were used to do the issue to help the classification in later work.This method can be generalized to the same optimizations of other kinds of biomedical tissues and even more applied fields.2.The leukemia blood cells in hyperspectral images were classified and counted roughly.Different from the traditional approaches used in 2D images,the classification of hyperspectral images could extract not only pictorial information,but also spectral features.In this thesis a classifier composed of artificial neural network and adaptive genetic algorithm was present to classify the leukemia blood cells.3.Pre-processing was achieved on hyperspectral images of retina slices of diabetic rats to improve the accuracy of quantitative analysis.Firstly,the format and characteristics of these images were introduced.Secondly,noise and stripes were removed.Finally,a statistical method was used to correct the system radiometric to deliminate the influence of light and CCD.4.The effect of EPO on retinopathy of diabetic rats induced chemically was evaluated preliminarily from the hyperspectral view.Since the investigation has been carried out by our cooperators through the traditional biochemical methods,the mission was taken with the help of hyperspectral images in this thesis.This technique may be an alternative way for pathologist to evaluate new medicine.