Identification on Adulteration of Rice Seeds by Terahertz Time£­ Domain Spectroscopy Based on Multi Feature Algorithm Selection

KeyWord:terahertz time-domain spectroscopy  pattern recognition  rice seeds  feature selection  adulteration identification

Abstract:

      A rice seed pattern recognition method based on terahertz time-domain spectroscopy was proposed in this paper in order to crack down on illegal elements peddling unqualified rice seeds in qualified seeds to obtain illegal profits£®Compared with traditional methods£¬the terahertz time-domain spectroscopy was fast£¬time-saving and non-destructive£¬which could meet the needs of front-line law enforcement personnel for rapid detection of samples£®In the experiment£¬10 kinds of rice seeds containing mixed adulteration of different brands were selected as samples£¬and the terahertz time-domain spectral data of the samples were collected£®Meanwhile£¬relief£¬random forest£¨RF£©£¬support vector machine recursive feature elimination£¨SVM£­RFE£© and maximum relation minimum redundancy£¨mRMR£© models were established to select the spectral wavelengths of the samples£¬respectively£®Finally£¬a classifier was designed to classify and identify the samples processed by the four feature selection methods£®The experimental results showed that the extreme learning machine£¨ELM£© model optimized based on the cuckoo search£¨CS£© algorithm had the best recognition effect on the sample spectral data extracted by the random forest feature selection algorithm£¬with an accuracy reaching 100%£®The experiment is of a certain reference significance for the identification of seed adulteration in the field of forensic science.