The Key Issue Research Of Distinguishing Corn Varieties’ Authenticity By Using Near Infrared Spectroscopy

The authenticity and purity of the crop seed is vital to food security and people’s livelihood. However,the seed market is not optimistic and need quickly and efficiently identify new methods. The commoncharacteristics of the varieties existing methods of seed identification is long detection cycle, high cost,Consuming labor and material resources. The methods cannot achieve the requirements of rapid, low-cost,easy to operate. Pattern recognition is an important part of artificial intelligence. Judgment and recognitionis the first step of intelligent machines to imitate human activities. Pattern recognition is the process ofprocessing, analyzing, judging and identifying various types of information of things or phenomena. NearInfrared Spectroscopy is a new technology of rapid detection of the chemical composition in one or moresamples, which uses the optical properties of organic chemicals in the near-infrared spectral region（wavelength780-2526nm）. With the combination with pattern recognition, near-infrared spectroscopy iswidely used in the testing of corn seeds. This paper analyzes the characteristics of the different bands of thenear-infrared spectroscopy in the identification of maize varieties, as well as the hazards and removal of thecoating on corn seeds.The detection of the corn varieties by near-Infrared Spectroscopy includes the following three steps:data acquisition, data analysis, data preprocessing; near-infrared spectral feature extraction; create arecognition model. Feature extraction methods include partial least squares, principal component analysis,linear discriminant analysis method, the minimum distance classifier, bio-mimetic pattern recognitionmethod. The multiple correlations existing between the data have a large impact on some of the commonlyused feature extraction methods such as principal component analysis. The multiple correlations of thevariables will distort the data in the number, so that the importance of some information is increased in thecalculation. So we need further improve some of the old methods to meet the needs of the experiment.The existing near-infrared spectroscopy-based species identification is the use of full-band or L-bandsignal and higher cost. Based on the combination of PCA, LDA, and BPR, a new corn varietiesidentification methods based on the short wavelength of the near-infrared spectroscopy is put forward. Byanalyzing the characteristics of the different bands of the near-infrared spectroscopy in the identification of maize varieties, we investigate the recognition rate of the short wavelength. For the Near-infrared spectraldata for the37maize varieties, we choose the starting wavelength as833nm and different cutoffwavelength, and then get different wavelength band data. For each band data, we respectively use （1）partial least squares regression prediction,（2）the PLS initial feature extraction, linear discriminant analysisof the second feature extraction, the minimum distance classifier to recognize,（3） principal componentanalysis to extract the initial characteristics, the second feature extraction in the LDA, the MDC to identify,（4） the PLS initial feature extraction, the LDA second feature extraction, pattern recognition methods toidentify,（5） PCA initial feature extraction, the LDA second feature extraction BPR recognition to identifyspecies. Experiments show that in the833-1087nm short wavelength the recognition rate of BPR methodreach to97.60%, the band narrowed down by84.71%and recognition rate fell by only2.16%comparedwith the full band. In the same case as the cut-off wavelength, the PCA feature extraction method and thePLS feature extraction method using the same classifier have different advantages and disadvantages andare greater than the PLS regression prediction method, but the recognition rate is not very different. Withthe cutoff wavelength getting shorter, the BPR methods using different feature extraction methods all canbe maintained at a high level, the identification of PLS regression prediction methods and the MDC methoddeclined sharply. Based on the above, we integrated the PCA LDA and BPR to propose a new corn varietiesidentifying method based on the short wavelength of the near-infrared spectroscopy. We proved byexperiments that using only the short wavelength of the near-infrared spectroscopy to identify corn varietiescan reach higher recognition rate.Near Infrared Spectroscopy is one of the main methods used to identify the authenticity of the cornseed, but the near-infrared light absorption of corn seed coating is very large, which have a big impact onthe test results. So the research on the corn seed coating is of great significance. After extensive research,we found that the coatings most make up of insecticide, fungicide, fertilizer, plant growth regulating agentand so on. The composition of pesticides, fungicides includes thiram C₆H₁2N₂S₄, carbofuran C₁2H₁5NO₃,tebuconazole C₁6H₂2CIN₃, imidacloprid C₉H₁0CIN₅O₂and triadimefon C₁4H₁6CIN₃O₂. These thingscontain hydrogen group, the absorption of infrared spectroscopy cannot be ignored. The maize seedcomposition is C, H, O, N and some trace elements. In order to remove the coating, we must first study thecoating to find its concentration and penetration depth. Based on the research of corn seed coating, we have chosen the S element as a characteristic element. S elements only exist in the corn coating, so we canindirectly measure the coating by detecting S element in the coating. By the XPS （X-ray photoelectronspectroscopy） we can achieve detection of various elements of the corn coating and corn seeds. Throughthe experimental apparatus, we can get the concentration of various elements of the different depth of theseed surface. According to the concentration change, we can determine the thickness of the coating tofacilitate subsequent experiments. The impressive results obtained by experiments on two different seeds. Selement concentration is reduced as the depth increases, which can be drawn from the S elementsdecreasing trend overall. So the thickness of the coating is limited and the thickness of the surface of thecorn seed is only at the level of micron.