Research On Similar Ellipsoid Chinese Herbal Medicine Image Recognition Algorithm Based Ongeneralized Multiple Kernel Learning

There are many different kinds of Chinese herbal medicine in our country.Some criminals often use substandard medicines fake rare medicinal materials,so that the Chinese herbal medicine market is filled with a large number of inferior product.At present, there is no complete classification standard of Chinese herbal medicine market.In order to improve Chinese herbal medicine market classification accuracy,China also have issued some policies to standardize the market for regulate Chinese medicine in recent years.In particular, computer vision technology applied widely in the field of traditional Chinese medicine classification,greatly enhance the objectivity and accuracy of identification of Chinese herbal medicines.In this paper,similarellipsoid Chinese herbal medicine image classification algorithm is studied. A set of basic algorithm is introduced from image database construction, image pre-processing, feature extraction and classification.Feature extraction of Chinese herbal medicineimage is focused on.The obtained characteristic value is used to classify the image of Chinese medicinal materials and to verify the classification effect.The main work of this paper includes following three aspects:(1)The image database of Chinese herbal medicine is constructed.In this paper, we select ten kinds of Chinese medicinal herbs in the market as the research objects. First, build a Chinese herbal medicine image acquisition system and captureimages.Secondly, thecapture image of the cropping, background segmentation and scaling.Finally, theimage libraryof ellipsoid Chinese herbal medicine was constructed after the pretreatment.This paper has collected Pinellia ternata, Pinellia ternate breit, Lyciumchinense, Lotus seeds, Ziziphus jujuba Mill and otherten kinds of Chinese herbal medicines. Each herbal medicine has hundredimages, a total of thousand images.(2)A scheme of image feature extraction is proposed. First of all, we select color histogram and color moment extraction image color feature,which focus on the color histogram of image color extraction and its improvement.Comparative results of the two classification algorithms, the improved color histogram can be more effective in extracting the color feature, and the recognition rate of the image isobviouslybetter than color moment feature extraction algorithm.Secondly, the local binary pattern(LBP)of the original texture feature extraction algorithm is improved. Through the fusion of Haar wavelet transform and LBP(HLBP)to classify Chinese herbal medicine image recognition.Experimental shows that classification performance of weighted HLBP(WHLBP)is better than the original LBP algorithm and HLBP algorithm. Finally, Using fourier descriptors and Hu invariant moments to extract image shape feature and carries on the experimental verification.Experimental results show that Fourier descriptors can be well expressed in medicine image shape information, image classification and recognition rate is higher than Hu moment.(3)The method of generalized multi-kernel learning is used to classifythe image of Chinese medicinal. The image color, texture and shape features of Chinese herbal medicines are extracted through mentioned above.Generalized multi-kernel learning method and Lib-SVM multi classification method are used to classify medicine image.Experimental results show that the generalized multiple kernel learning method can obtain ideal results.This section focuses on generalized multi-kernel learningof influence of parameters on the effect of Chinese herbal medicine image classification, and get the best recognition rate of optimal parameters of generalized multi-kernel learning method.