Comparative Study Of Machine Learning Classification Of Yeast Microscopic Images And Diffraction Images

Yeast germination rate is an important indicator of yeast activity.Microscopic analysis can be used to visually determine the germination of yeast.In this dissertation,a new method for yeast germination identification based on flow cell diffraction image and machine learning technology is proposed,and the original polarization diffraction imaging flow cytometer(p-DIFC)is modified to obtain the microscopic image and diffraction image of yeast simultaneously.The main work of this topic include the followings:1)Optical improvement of the p-DIFC: The main improvement of the optical path is to change the P direction into an infinite microscopic optical path with about80 X magnification based on the combination of the objective lens and the tube lens,.2)Acquisition and preprocessing of yeast images: Yeast solution is prepared by the dry yeast on the market,and tested by the improved diffraction imaging flow cytometer for acquisition of microscopic images and diffraction images simultaneously.The underexposed and overexposed images are removed in the preprocessing step.3)Artificial labeling of yeast image: The images of the yeast are labeled with the microscopic image as a standard into three types: budding,non-germination,and others(agglomeration of three or more yeasts).4)Comparative study of machine learning classification methods for yeast image : Two methods were used to classify yeast images.The first method is based on gray level co-occurrence matrix(GLCM).Firstly,the labeled data is divided into training set and test set,and extract the selected GLCM feature parameters.The support vector machine and random forest are used to train the model,verified with a simple cross-validation method.The second method is based on the Convolutional Neural Network(CNN)method,which used the established model with 8 convolution layers to train and validate the model.The results show that the two methods have achieved 96% and 95% accuracy in the classification of diffraction images,respectively.The classification accuracy of GLCM + SVM(Support Vector Machine)and CNN algorithm based on diffraction images is significantly better than that of ordinary microscopic images(P < 0.01).The above results confirm the validity of diffraction imaging flow cytometry for identification of yeast germination,and provide a new example for further application of diffraction imaging flow cytometry.