| ObjectiveTo establish a stitching method of the transverse section microscopic images of Achyranthes Bidentatae Radix,Cyathulae Radix,Saposhnikovia Radix and to extract the tissue distribution features of materials.To extract the morphological characteristics from images of Achyranthes Bidentatae Radix,Cyathulae Radix,Saposhnikovia Radix.Combined with a variety of digital image processing technology to establish the identification system of traditional Chinese herbal medicines of tissue distribution features and morphological characteristics,it can achieve the Pattern recognition of images of traditional Chinese herbal medicines and further provide new ideas and methods for the identification of traditional Chinese herbal medicines.MethodsThe tissue feature extraction and recognition method:The medicinal materials were taken in the middle of the sample and then made into transverse section by embedding to the polyethylene glycol.Using microscope and electron eyepiece to shoot the microscopic image of whole cross section in order.Using a hierarchical search strategy to improve the precision and speed of image stitching on the basis of block matching method in image registration and using wavelet fusion algorithm to improves image fusion effect on the basis of the average gray adjustment in image fusion,it can complete the microscopic image stitching of the traditional Chinese Medicine on the basis of Matlab.Using the improved Sobel operator based on four direction edge detection and the mathematical morphology operation to the microscopic images of Segmentation and extraction,Using the BP neural network,Bayesian classification and k-nearest neighbor classification to pattern recognition of features.The morphological characteristic extraction and recognition method:Using camera to take pictures of medicinal materials,Using the best threshold segmentation method based on HSV color space for s-components to the medicinal images of Segmentation and extraction,Using the BP neural network,Bayesian classification and k-nearest neighbor classification to pattern recognition of features.Finally,based on the Matlab programming platform,GUI is built with the tissue characteristics and character of the medicinal herbs were extracted from the image to the pattern recognition.Results(1)A stitching method of the transverse section microscopic images of Achyranthes Bidentatae Radix,Cyathulae Radix,Saposhnikovia Radix was established.(2)We developed the microscopic image mosaic GUI of Achyranthes Bidentatae Radix,Cyathulae Radix,Saposhnikovia Radix on the basis of Matlab,and the pattern recognition GUI of Chinese herbal medicine on the basis of Matlab.(3)Using the k-nearest neighbor classification,Bayesian classification and BP neural network to pattern recognition of features.In terms of organizational characteristics,with randomly select different training samples,The average recognition rates of three pattern recognition methods were calculated as 93.8%,98.2%and 94.7%,respectively(n = 75).In terms of morphological characteristics,with randomly select different training samples,The average recognition rates of three pattern recognition methods were calculated as 64.2%,91.78%and 96.22%,respectively(n = 150).Based on the combining organizational characteristics and morphological characteristics of randomly select different training samples,the average recognition rates of three pattern recognition methods were calculated as 91.6%,99.6%and 94.7%,respectively(n = 75).In the experiment,the Bayesian classification method had the highest recognition rate(99.6%)in terms of samples of combine organizational characteristics and morphological characteristics.ConclusionsIn microscopic image stitching,a hierarchical search strategy can improve the precision and speed of image stitching on the basis of block matching method in image registration;while wavelet fusion algorithm can enhance image fusion effect on the basis of the average gray adjustmen in image fusion.Thus,the aforementioned two technologies can be used for the stitching of the transverse section microscopic images of Chinese herbal medicine.The simple and fast feature-extraction method can be applied for a variety of image segmentation and characterization identification of Chinese herbal medicines.The comparion of the recognition results in different pattern recognition methods towards different Chinese herbal medicines,and the highest recognition rate using the Bayesian classification method in terms of Chinese herbal medicine of combine organizational characteristics and morphological characteristics,thus can provide the basis for the automatic identification method of Chinese herbal medicine.Establishment of the automatic identification of Chinese herbal medicine GUI is a kind of graphical user interface;The method can eliminat the cumbersome program code processing,and is considered as a more convenient one for the automatic identification of Chinese herbal medicines. |
PDF Full Text Request