Cowpea (Vigna unguiculata L. Walp) is a multi-purpose legume with high quality protein for human consumption and livestock. The objective of this work was to develop near-infrared spectroscopy (NIRS) prediction models...Cowpea (Vigna unguiculata L. Walp) is a multi-purpose legume with high quality protein for human consumption and livestock. The objective of this work was to develop near-infrared spectroscopy (NIRS) prediction models to estimate protein content in cowpea. A total of 116 cowpea breeding lines with a wide range of protein contents (19.28 % to 32.04%) were selected to build the model using whole seed and ground seed samples. Partial least-squares discriminant analysis (PLS-DA) regression technique with different pre-treatments (derivatives, standard normal variate, and multiplicative scatter correction) were carried out to develop the protein prediction model. Results showed: 1) spectral plots of both the whole seed and ground seed showed higher spectral scatter at higher wavelengths (>1450 nm), 2) data pre-processing affects prediction accuracy for bot whole seed and ground seed samples, 3) prediction using ground seed samples (0.64 R<sup>2</sup> 0.85) is better than the whole seed (0.33 R<sup>2</sup> 0.78), and 4) the data pre-processing second derivative with standard normal variate has the best prediction (R<sup>2</sup>_whole seed = 0.78, R<sup>2</sup>_ground seed = 0.85). The results will be of interest in cowpea breeding programs aimed at improving total seed protein content.展开更多
文摘Cowpea (Vigna unguiculata L. Walp) is a multi-purpose legume with high quality protein for human consumption and livestock. The objective of this work was to develop near-infrared spectroscopy (NIRS) prediction models to estimate protein content in cowpea. A total of 116 cowpea breeding lines with a wide range of protein contents (19.28 % to 32.04%) were selected to build the model using whole seed and ground seed samples. Partial least-squares discriminant analysis (PLS-DA) regression technique with different pre-treatments (derivatives, standard normal variate, and multiplicative scatter correction) were carried out to develop the protein prediction model. Results showed: 1) spectral plots of both the whole seed and ground seed showed higher spectral scatter at higher wavelengths (>1450 nm), 2) data pre-processing affects prediction accuracy for bot whole seed and ground seed samples, 3) prediction using ground seed samples (0.64 R<sup>2</sup> 0.85) is better than the whole seed (0.33 R<sup>2</sup> 0.78), and 4) the data pre-processing second derivative with standard normal variate has the best prediction (R<sup>2</sup>_whole seed = 0.78, R<sup>2</sup>_ground seed = 0.85). The results will be of interest in cowpea breeding programs aimed at improving total seed protein content.