Research Of Single Cell Image Feature Extraction And Recognition Based On The Time-frequency Analysis Method

Currently.cell detection technology based on manual interpretation exits a large amount of work,high cost,the reliability and accuracy of the detection affected by physician subjective and other problems.With the rapid development of computer technology,digital image processing technology and artificial intelligence pattern recognition technology, it is urgent to develop micro cell intelligence analysis system. It is very important to realize the micro cell intelligence analysis system, which has very important social and economic benefits.In this paper,the key steps of single cell image intelligent recognition are intensive studied,including the single cell image preprocessing,single cell image feature extraction, single cell image feature optimization and single cell image recognition.Time-frequency analysis method is introduced into the single cell image processing,establishing the single cell recognition model based on the time-frequency analysis method and the compressed sensing theory.The main research results are as follows:1.A preprocessing method based on wavelet parameter optimization for single cell image denoising is proposed,for removing the disturbance in a single cell image and preparing for the subsequent cell recognition.We propose improvement for the traditional wavelet thresholding denoising based on he wavelet transform multi-scale and multi-resolution feature.(1)Compared to traditional single threshold, the threshold estimation method of an adaptive scale parameter value is proposed.(2)To overcome the shortcomings of the soft and hard threshold and wavelet coefficients in less than the threshold being set zero directly,construct a new threshold function to enhance wavelet denoising effect.In addition, this paper also introduces the evaluation method of the denoising effect.The effectiveness of the proposed method is verified by the cervix single cell image.2.Single cell image feature extraction method based on wavelet transform and two-dimensional empirical mode decomposition is proposed. Wavelet transform and two dimensional empirical mode decomposition has the fine features of multi-scale and multi-resolution,which overcomes the shortcomings of the single-scale feature extraction in spatial domain. Wavelet transform and two-dimensional empirical mode decomposition method can not only extract the multi-scale frequency domain characteristics of single cell images,but also can extract the phase characteristics of single cell images,combined with the singular value decomposition method, can obtain the effect of larger inter-classr variance and smaler intra-class variance of single cell image feature. The method presented in this paper can fully extract the features of single cell images and the intrinsic structure and texture features of single cells. The validity of this method is verified on a single cell data set of the cervix.3.The single cell image features are optimized by the method of weed optimization. The feature of single cell image extraction is redundant, and the high dimension feature is a great burden to the classifier,.effecting the single-cell recognition accuracy and recognition time.Therefore, this paper introduces the weed optimization algorithm for optimizing the high dimensional feature of single cell image.The effectiveness of the proposed method is verified on a single cell data set of the cervix, and the recognition effect with optimization feature is better than without optimization feature.4 A single cell image recognition model based on compressive sensing theory is established.The compressed sensing classifier is representing the feature vector of uncategorized single cell imagesample as a linear combination of the feature vector of the single cell image training sample,determine the corresponding single cell category depend on the coefficient.The single cell image recognition model based on compressive sensing theory is a new method for the fast blind identification of single cell image without the need for a single cell image segmentation and precise positioning. The compressed sensing classifier makes up for the shortcomings of the low recognition rate and the long time consuming of the existing classifiers. The single cell image recognition model is verified on the single cell image data set of the cervix, and the results show that the compressed sensing classification has good real-time performance, robustness and recognition performance.5.A single cell image processing and recognition system software is developed based on MATLAB GUIplatform, which realizes the visualization and interactive dynamic effect of single cell image processing. The single cell image processing software based on MATLAB GUI has strong practicability and application value in the field of medical image processing.