Research On The Identification Method Of Aucklandiae Radix Based On Machine Olfaction

Aucklandiae radix is a kind of herbal medicine of Compositae family.It is a good medicine for treating stomachache and diarrhea and accompanied by strong aroma.For aucklandiae radix collected at different origin or different harvesting time,because of the different conditions of light,moisture,temperature and nutrient during cultivation,their major function is different,which affects their sale prices.Different kinds of aucklandiae radix are very similar in appearance,and it is difficult for ordinary consumers without special training to distinguish their types from the appearance.This paper proposes an identification method of aucklandiae radix based on machine olfaction.By analyzing and processing the odor information of aucklandiae radix from different origin or different harvesting time,this method can identify the species of aucklandiae radix and provide a new detection method for aucklandiae radix.Machine olfaction can simulate the working mode of biological olfaction by computer.It consists of the gas sensor array and the odor signal processing system.Traditional methods of machine olfaction mainly deal with odor information by linear projection.The emerging manifold learning has introduced a new solution to machine olfaction,which can embed highdimensional data into low-dimensional space and capture the non-linear characteristics of the original data better.In this paper,a new machine olfaction method called t-SNE+LDA is proposed by combining a manifold learning method called t-SNE with linear discriminant analysis.This method maps high-dimensional aucklandiae radix data to low-dimensional space by t-SNE,and then classifies and identifies them by using classifier based on linear discriminant analysis.t-SNE can effectively alleviate the common "crowding problem" in dimensionality reduction algorithm and obtain better anti-noise ability,and it can retain the local and global structural features of the original data.In order to analyze aucklandiae radix from two perspectives of origin and harvesting time,aucklandiae radix from different origin and different harvesting time was used as experimental samples.The odor data of these samples was collected by an electronic nose.The collected odor data was sorted out and processed by imitating the storage form of general dataset,and the dataset of aucklandiae radix odor information was formed.In order to verify the effect of t-SNE+LDA on the classification and identification of aucklandiae radix odor data,four algorithms,SLLE,SLLE+LDA,t-SNE and t-SNE+LDA,were applied in the experiment.The results of these algorithms were compared through the angles of classification effect and identification accuracy.(1)It can be found that t-SNE can provide better dimension reduction results for LDA classification by comparing the experimental results of SLLE+LDA and t-SNE+LDA,(2)By comparing the experimental results of SLLE and SLLE+LDA,t-SNE and tSNE+LDA,it can be found that LDA projected the data to a better classification space after the dimensionality reduction of SLLE or t-SNE,and obtained a better discriminant effect.Experiments show that t-SNE+LDA combines the advantages of the two algorithms,and has better classification results with larger inter-class scatter and smaller intra-class scatter,and has higher identification accuracy.It is an effective method to identify odor data of aucklandiae radix.