Design And Classification Application Of Microscopic Hyperspectral Imaging System Based On LCTF

Hyperspectral imaging technology combines the advantages of spectral analysis and traditional imaging technology.It integrates many disciplines such as optics,computer technology and electronic technology.It is widely used in crop maturity prediction,food safety detection,atmospheric prediction and defense military fields.The microscopic hyperspectral imaging system consists of a microscope and a spectral imaging system that can acquire hyperspectral images of tiny substances.It has played an important role in the field of life sciences,especially cytology,and has attracted the attention of many scholars.This paper improves the structure of microscopic hyperspectral imaging system based on liquid crystal tunable filter.Adopt corresponding preprocessing measures for the acquired cell spectral image data set.And a feature extraction method combining texture features and visual bag-of-words model is proposed.In this paper,we use weighted support vector machine and convolutional neural network to carry out cell spectral image classification research,verifying the advantages of spectral image in the field of cell recognition.The main research contents and results of this article are as follows:(1)A brief introduction to the structure of the laboratory's self-organized microscopic hyperspectral imaging system and the working principle of each component.The spectroscopic device selects a liquid crystal tunable filter with low energy consumption and less stray light.The hardware parameters of the system are introduced and the performance of the system is analyzed.Use network flow to transfer data between the industrial computer and Smart DC to increase system flexibility.Through system debugging experiments,the system parameters of the collected cell spectral data set were determined,including camera frame rate,exposure time,and wavelength queue interval.(2)The self-organized microscopic hyperspectral imaging system is used to collect 5 types of cell spectral images,with a spectral range of 420?730nm and a spectral resolution of 5nm.Each cell spectral image contains 63 bands.In order to eliminate the influence of environmental factors and system optics on the spectral image,the image is pre-processed using the black and white correction method.Four texture feature parameters are selected to extract the texture features of the image,and the bag of visual words(BOVW)model is used for feature extraction.A featureextraction method combining the two is proposed.The spectral angles and spectral distances of the five types of images are quantitatively analyzed to verify the advantages of spectral images for cell recognition.(3)Experimental research on classification of collected cell spectral data set and Indian Pines dataset.The weighted SVM classification model and the CNN classification model were used for classification experiments,and the JM distance was introduced to weight the kernel function to make full use of the spectral dimension information.The CNN model can either complete feature extraction or directly output classification labels,eliminating the design process of feature extraction.The above experimental research shows that it is feasible to combine microscopic hyperspectral imaging technology and pattern recognition algorithm for cell recognition.