牛膝、川牛膝、防风各组织、性状的特征提取与模式识别

发布时间：2018-12-13 02:46

【摘要】：目的:建立中药材牛膝、川牛膝、防风横切面显微图像的拼接方法,提取中药材的组织分布特征。由牛膝、川牛膝、防风药材图像,提取中药材的性状特征。结合多种数字图像处理技术,建立以中药材的组织特征、性状特征为参数的药材鉴定系统,实现中药材图像的模式识别,为中药鉴定的研究提供新思路、新方法。方法:组织特征提取、识别方法:取药材样品中段,以聚乙二醇包埋制片法制成横切片。用显微镜和电子目镜依次拍摄整个横切片的显微图像。以Matlab为编程平台,采用基于金字塔式分层搜索策略的块匹配法进行图像配准,用基于直接平均灰度调整的小波融合算法进行图像融合,完成药材的显微图像拼接。运用基于改进的四个方向边缘检测sobel算子和数学形态学开闭运算进行显微图像的分割、特征提取,用BP神经网络法、贝叶斯分类法、k-近邻分类法进行特征的模式识别。性状特征提取、识别方法:采用相机拍摄药材的图像,以Matlab为编程平台,采用基于hsv颜色空间的s分量最佳阈值分割法进行药材图像的分割、特征提取,用BP神经网络法、贝叶斯分类法、k-近邻分类法进行特征的模式识别。最后构建基于Matlab编程平台的药材组织特征、性状特征从图像提取到模式识别的GUI。结果:(1)建立了中药材牛膝、川牛膝、防风横切面显微图像的拼接方法。(2)建立了基于Matlab编程平台的牛膝、川牛膝、防风不同药材显微图像拼接的GUI以及中药材模式识别的GUI。(3)采用k-近邻分类法、贝叶斯分类法、BP神经网络法识别药材特征。在组织特征上,随机选取不同的训练样本,三种模式识别方法对药材的平均识别率分别为93.8%、98.2%、94.7%(n = 75);在性状特征上,随机选取不同的训练样本,三种模式识别方法对药材的平均识别率分别为64.2%、91.78%、96.22%(n= 150);同时在综合组织特征和性状特征上,随机选取不同的训练样本,三种模式识别方法对药材的平均识别率分别为91.6%、99.6%、94.7%(n = 75)。实验中采用贝叶斯分类法对综合药材组织特征与性状特征的样本识别率最高(识别率=99.6%)。结论:显微图像拼接中采用基于金字塔式分层搜索策略的块匹配进行图像配准,用基于直接平均灰度调整的小波融合算法进行图像融合既保证了图像拼接的速度,又保证了图像拼接的精度,可用于中药材横切面显微图像的拼接。选用的特征提取方法简便、快捷,适合多种图像的分割以及中药材鉴定特征的描述。比较了多种模式识别方法在不同药材特征上的识别结果,采用贝叶斯分类法识别药材的性状特征与组织特征获得的识别率最高,为中药材自动化识别的方法选择提供依据。构建的中药材自动化识别GUI是一种人机交互的图形用户界面,省去了繁琐的程序代码处理,为中药材的自动化识别提供了更加方便、快捷的操作。
[Abstract]:Objective: to establish a method of stitching the microimages of Radix Achyranthes bidentata, Radix Achyranthes bidentata and Radix Astragali on cross section, and to extract the tissue distribution characteristics of Chinese medicinal materials. The characters of Chinese medicinal materials were extracted from the images of Radix Achyranthes bidentata, Radix Achyranthes bidentata and Radix Fangfeng. Combined with various digital image processing techniques, a medicinal identification system based on the tissue and character characteristics of Chinese medicinal materials was established to realize pattern recognition of traditional Chinese medicine images and to provide new ideas and methods for the study of traditional Chinese medicine identification. Methods: the method of tissue feature extraction and identification was as follows: the middle part of the sample was prepared by polyethylene glycol embedding method. Microscope and electronic eyepiece were used to photograph the whole transverse slice in turn. Using Matlab as the programming platform, the block matching method based on pyramid hierarchical search strategy is used for image registration, and the wavelet fusion algorithm based on direct average gray level adjustment is used for image fusion to complete the microscopic image mosaic of medicinal materials. The sobel operator based on improved four-direction edge detection and mathematical morphology are used to segment and extract the microscopic image. The pattern recognition is carried out by BP neural network method Bayesian classification method and k-nearest neighbor classification method. The method of character feature extraction and recognition: taking the image of medicinal materials by camera, taking Matlab as the programming platform, using the best threshold segmentation method of s component based on hsv color space to segment the image of medicinal materials, extracting the feature, using BP neural network method. Bayesian classification and K-nearest neighbor classification are used for feature recognition. Finally, a GUI. based on Matlab programming platform is constructed for the tissue features of medicinal materials and character features from image extraction to pattern recognition. Results: (1) the methods of stitching the microimages of Radix Achyranthes bidentata, Radix Achyranthes bidentata and Radix angelicae officinalis were established. (2) the Matlab programming platform was established for the mosaic of the images of Achyranthes bidentata and Radix Achyranthes bidentata. GUI of different medicinal materials and GUI. (3) of pattern recognition of Chinese medicinal materials were identified by using k- nearest neighbor classification, Bayesian classification and BP neural network method. In terms of tissue characteristics, different training samples were randomly selected. The average recognition rate of the three pattern recognition methods for medicinal materials was 93.898. 2% (n = 75). In terms of traits, different training samples were randomly selected. The average recognition rate of the three pattern recognition methods was 91.78% (n = 150), and the average recognition rate was 91.78% (n = 150). At the same time, the average recognition rate of the three pattern recognition methods for medicinal materials was 91.6% and 99.6% (n = 75), respectively, on the basis of the comprehensive tissue and character characteristics, and different training samples were selected randomly. The average recognition rates of the three pattern recognition methods were 91.6% and 99.6% (n = 75), respectively. In the experiment, Bayesian classification method was used to identify the samples with the highest recognition rate (99.6%). Conclusion: in microscopic image mosaic, block matching based on pyramid hierarchical search strategy is used for image registration, and wavelet fusion algorithm based on direct average gray level adjustment can ensure the speed of image stitching. The precision of image stitching is guaranteed, and it can be used in cross-section micro-image of Chinese herbal medicine. The selected feature extraction method is simple, fast and suitable for segmentation of various images and description of identification features of Chinese medicinal materials. The results of different pattern recognition methods in different medicinal materials were compared. Bayesian classification was used to identify the characters and tissue characteristics of medicinal materials, which provided the basis for the automatic identification of Chinese medicinal materials. The automatic identification GUI is a graphical user interface with human-computer interaction, which eliminates the tedious program code processing, and provides a more convenient and quick operation for automatic identification of Chinese medicinal materials.
【学位授予单位】：广州中医药大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：R282.5;TP391.41

【相似文献】