Rapid And Nondestructive Analysis For The Identification Of Rice Seeds With Near Infrared Spectroscopy

With the development of society,the demand for high-quality foods is increasing.Rice is one of the important crops.The authenticity of seeds is the key to rice yield and quality.The seed varieties objectively mixed in the breeding process,or subjectively shoddy that will seriously affect the yield and quality of rice cultivation.In addition it is also not allowed by the national seed law.Therefore,seeds identification is essential.However,the existing manual methods are backward and the accuracy is low.And some other laboratory methods are complicated,time-consuming and expensive.So there is an urgent need for a simple,rapid and effective detection technology.Near-infrared spectroscopy is a fast,simple and green analysis technique that has been applied to the non-destructive and rapid analysis of major components of agricultural products.However,researches on seeds identification by near-infrared spectroscopy are still rare,especially for the accurate discriminant analysis of seeds mixture.Due to the complexity of the problem,it has not yet been resolved.In this study,156 portions of Yueliangyou-165(R1)were used as identification samples(negative).And the 156 interference samples(positive)were comprised of 15 portions of Fuliangyou-2168(R2),15 portions of Nanguizhan(R3),15 portions of Nanjingzhan(R4),15 portions of Yuejinyouzhan(R5),15 portions of Yueyousimiao(R6)and 81 portions of composite samples with different proportions of R1 mixing with other variety seeds.(1)Method research 1)standard normal variate(SNV)was used as a spectral pre-processing method;partial least squares discriminant analysis(PLS-DA)was used as a spectral pattern recognition method;moving window(MW)and equidistant combination(EC)were used as wavelength selection methods;the integrated optimization methods for spectral discriminant analysis were proposed,such as SNV-MW-PLS-DA and SNV-EC-PLS-DA.2)A more concise bis-correlation coefficients(BiCC)was proposed as a spectral pattern recognition method;the integrated optimization method for spectral discriminant analysis were further proposed,such as SNV-MW-BiCC and SNV-EC-BiCC.3)Furthermore,an equivalent optimal model space and its simplification method were proposed.(2)Experimental research 1)Construct a sample partitioning system for calibration,prediction and inspection: 96 samples were randomly selected for testing,and the remaining 216 samples were randomly divided(20 times)into calibration(120)and prediction(96).Parameters optimization was performed in the modeling process;during the testing process,samples which were not involved in modeling were used to test the established model.2)Compared with other methods,the SNV-EC-PLS-DA method achieved the best modeling effect.The total recognition rate of modeling and testing reached 100% and 98% respectively.The equivalent optimal model space contains 25 wavelength models,which were concentrated in the spectral region(750-1100 nm)which connects visible and short-wavelength near-infrared region.3)The total recognition rate of modeling and testing of SNV-EC-BiCC method reached 98% and 97%,respectively.The results indicated that near-infrared spectroscopy combined with SNV-EC-PLS-DA,SNV-EC-BiCC and other methods with high-accuracy and non-destruction in discriminant analysis can be applied for the distinction of rice seeds.Equidistant combination wavelength selection method can effectively eliminate redundant wavelengths,extract information wavelengths,and reduce model complexity.The proposed band model can be used for the design of small-scale spectrometers.The proposed analysis method is simple,rapid and effective,and has important application prospects in the field of rice seeds detection.