Vis-NIR Wavelength Optimization For Non-Destructive Discriminant Analysis Of Transgenic Sugarcane Leaves

Based on Savitzky-Golay(SG) spectral pre-treatment method, a coupling model of principal component(PCA) and linear discriminant analysis(LDA) is combined with waveband screening strategy of moving window(MW) and equidistant combination(EC). Two integrated optimizing methods for spectral pattern recognition, MW-PCA-LDA and EC-PCA-LDA, are developed and successfully employed for the non-destructive recognition of transgenic sugarcane leaves via visible(Vis) and near-infrared(NIR) diffuse reflectance spectroscopy. On the basis of spectral experiment and computer optimization experiment with large samples, through a process of calibration, prediction and validation, two discrimination models with high accuracy are established.A total of 456 samples of sugarcane leaves in elongating stage are collected from a planted field. These samples are composed of 306 transgenic samples containing both bacillus thuringiensis(Bt) and biolaphos resistance(Bar) genes, which are determined via Enzyme-linked Immuno Sorbent Assay(ELISA), the remaining 150 are non-transgenic samples. The Vis-NIR spectral for all the samples are collected(400-2498 nm) and used for calibration, prediction and validation. According to the discriminant accuracy for prediction, the optimal SG parameters d, p, m are 1, 3, 25, respectively. Furthermore, For SG-MW-PCA-LDA method, the optimal waveband is 768 nm to 822 nm, the optimal PC combination(PCC) is PC1-PC3 and the corresponding validation recognition rates of transgenic and non-transgenic samples achieve 99.1% and 98.0%, respectively. For SG-EC-PCA-LDA method, the optimal wavelength combination contains only five points, which are 766 nm, 820 nm, 874 nm, 928 nm and 982nm; the optimal PCC is PC1-PC3 and the corresponding validation recognition rates of transgenic and non-transgenic samples achieve 100% and 98.0%, respectively.The results show that Vis-NIR spectroscopy can be used for accurate discrimination of transgenic sugarcane leaves. The proposed spectral pattern recognition methods(SG-EC-PCA-LDA, SG-EC-PCA-LDA) can effectively eliminate the noise within spectrum, extract information and acquire highly accurate discrimination effects. These rapid and convenient methods can provide a potential means for screening transgenic sugarcane breeding in large-scale agricultural production.