The Inspection Methods Of Feed Main Nutritional Components By Nirs And Hyperspectral Imaging

Detection of feed nutrient composition is one of the important technical means to ensure the quality of compound feed products. In order to research the rapid detection methods of near infrared reflectance spectroscopy(NIRS) and hyperspectral imaging,which are used to detect the nutrition components of the compound feed, 403 samples of compound feed have been collected. The near- infrared spectroscopy and visible /near infrared spectroscopy has been respectively obtained by using near infrared spectrometer and hyperspectral imager. Combined with methods of eliminating outliers、dividing sample set、optimal spectral pretreatment methods and selection of the optimal optical wave length, we have used partial least squares(PLS) stoichiometric methods to respectively establish the quantitative analysis model of crude protein, crude ash, moisture, total phosphorus, calcium content in compound feed based on near infrared spectrum and hyperspectral image technology. The main research contents and results are as follows.1) The quantitative analysis models of crude protein, crude ash, moisture,total phosphorus and calcium content in compound feed which is based on near infrared spectroscopy has been established. Collected infrared reflectance spectroscopy information of samples and used leverage- students residuals to eliminate outliers.The SPXY is the best partition method for crude protein,and C G is the optimal method for crude ash、water、total phosphorus and calcium content. The optimization spectra pretreatment method of each primary nutrients has been found, correlation coefficient method has been used to select the optimal optical wave length.The determination coefficient of calibration set(R2c) of crude protein is 0.8623, the root mean square error of calibration(RMSEC) is 2.0718%,the relative error analysis of calibration set(RPDc) is 2.5582, the determination coefficient of validation set(R2v) is 0.8001,the root mean square error of prediction(RMSEP) is 1.8709 %,the relative error analysis of validation set(RPDv) is 2.9558. The R2 c of crude ash is 0.7198, RMSEC is 1.1765%, RPDc is 1.8954, R2 v is 0.8039, RMSEP is 1.0050%, RPDv is 2.2388. The R2 c of water is 0.7446, RMSEC is 1.5546%, RPDc is 1.9789, R2 v is 0.6869, RMSEP is 1.7131%, RPDv is 1.7921. The R2 c of total phosphorus is 0.8059, RMSEC is 0.1166%, RPDc is 2.3156, R2 v is 0.7876, RMSEP is 0.1178%, RPDv is 2.2071.The R2 c of total phosphorus is 0.8059, RMSEC is 0.1166%, RPDc is 2.3156, R2 v is 0.7876, RMSEP is 0.1178%, RPDv is 2.2071.The R2 c of calcium is 0.3270, RMSEC is 0.1903%, the RPDc is 1.2086, R2 v is 0.2766, RMSEP is 0.1978%, RPDv is 1.1628.The results showed that, the quantitative analysis model of crude protein, crude ash content and total phosphorus showed better estimated performance; quantitative analysis model of water prediction accuracy is still not ideal, remains to be further optimized; while the quantitative analysis model of calcium prediction ability is poor, cannot be used for quantitative analysis.2) The quantitative analysis models of crude protein, crude ash, moisture, total phosphorus and calcium content in compound feed which is based on hyperspectral image has been established. Collected visible/infrared reflectance spectroscopy information of samples and used leverage- students residuals to eliminate outliers.The SPXY is the best partition method for crude protein,and C G is the optimal method for crude ash、water、total phosphorus and calcium content. The optimization spectra pretreatment method of each primary nutrients has been found, correlation coefficient method has been used to select the optimal optical wave length.The R2 c of crude protein is 0.8373, RMSEC is 2.1327%, RPDc is 2.4851, R2 v is 0.7778, RMSEP is 2.6155%, RPDv is 2.1143.The R2 c of crude ash is 0.7932, RMSEC is 1.0107%, RPDc is 2.2064, R2 v is 0.7758, RMSEP is 1.0611% RPDv is 2.1204. The R2 c of water is 0.6470, RMSEC is 1.8221%, RPDc is 1.6849, R2 v is 0.6314, RMSEP is 1.6003%, RPDv is 1.9371. The R2 c of total phosphorus is 0.6038, RMSEC is 0.1656%, RPDc is 1.5700, R2 v is 0.4672, RMSEP is 0.1916%, RPDv is 1.3570.The R2 c of calcium is 0.4784, RMSEC is 0.1676%, RPDc is 1.3723, R2 v is 0.4406, RMSEP is 0.1755%, RPDv is 1.3105.The results showed that, the quantitative analysis model of crude protein, crude ash content showed better estimated performance; quantitative analysis model of water prediction accuracy is still not ideal, remains to be further optimized; while the quantitative analysis model of calcium and total phosphorus prediction ability is poor, cannot be used for quantitative analysis3)The PLS quantitative analysis model of those nutrient composition was established in the optimal optical wave length based on near infrared reflectance spectroscopy and hyperspectral imaging technology.By comparing those models we found that the effect of the optimal model of crude protein established by the NIRS is more desirable. The determination coefficient of calibration set reach to 0.8623, which can be used in the actual quantitative analysis.Besides, both the quantitative analysis model of crude ash based on NIRS and hyperspectral imaging technology can be used in the actual quantitative analysis.However, the quantitative analysis model based on hyperspectral imaging technology is more accurate than the quantitative analysis model based on NIRS technology in prediction.Although,the prediction effect of the quantitative analysis model of water established on NIRS technology is better than the model established on hyperspectral imaging technology, the accuracy of the prediction of it is still unsatisfied.The model of total phosphorus based on NIRS performed well in prediction,which can be used in the actual detection.The analysis model of total phosphorus、calcium established on the hyperspectral imaging technology as well as the quantitative analysis model of calcium established on NIRS performed unsatisfied in prediction, both of them cannot be used in actual detection.