Construction Of Fast And Nondestructive Determination Model Of Protein,Oil,and Gossypol Contents In Cottonseeds Using Near Infrared Spectroscopy

Cotton is an important natural plant fiber crop and oilseed crop in China.Cottonseed,an important by-product produced in the process of cotton production,is rich in nutritional components.Cottonseed protein is rich in essential amino acid and cottonseed oil is rich in essential fatty acid.However,it is gossypol,which hinder their application.The contents of protein,oil and gossypol are important indexes to evaluate the quality traits of cottonseed.Usually,the determination of these contents is performed according to accurate and sensitive chemical methods,but the methods need tedious pretreatment and are relatively time consuming and expensive.In recent years,the rapid non-destructive and cheap near infrared spectroscopy?NIRS?technique has been taken more and more attention.This experiment was conducted to establish build a fast and nondestructive determination method of crude protein content,oil and gossypol content in cottonseed kernel.It will be of great importance for the selection and breeding of cotton which contains high protein,high oil,and low gossypol and the rapid detection and evaluation of nutritional quality of cottonseed.The contents of crude protein,oil,gossypol and fatty acid of 176 representative samples were evaluated.The near-infrared spectra of hairy cottonseed were collected by the Foss near infrared fast component analyzer.The near-infrared calibration models were developed by using a modified partial least square?MPLS?regression method as well as different math and scatter treatment methods in the range of optimum wavelength.In order to test the predictability and stability of the model ulteriorly,the established calibration model was used to predicted prediction set.The main research results are as follows:?1?The mean crude protein content of cottonseed kernels was 47.68%,ranging fr om 42.11%to 53.34%.The majority of them are belonging to 45%-51%level.The m ean oil content of cottonseed kernels was 27.56%,ranging from 22.70%to 34.08%.T he majority of them are belonging to 25.5%-29%level.The mean gossypol content of cottonseed kernels was 0.45%,ranging from 0.19%to 0.74%.The majority of them are belonging to 0.3%-0.6%level.Cottonseed oil contained thirteen fatty acids.The m ain compounds of them were linoleic acid?51.99%-60.88%?,palmitic acid?18.30%-25.68%?,and oleic acid?12.28%-18.50%?.Unsaturated fatty acids and essential fatty acid s constituted 73.62%and 57.44%of total fatty acids,respectively.Myristic acid,palmi tic acid,palmitoleic acid,stearic acid,oleic acid,malvalic acid,linoleic aicd,sterculic acid,dihydrosterculic acid and arachidic acid were the characteristic constituents of th e fatty acids in cottonseed oil.?2?Near-infrared spectroscopy model testing crude protein content:the determination coefficient of cross validation?1-VR?was 0.932,the determination coefficient of calibration(RSQ_cal)was 0.952,the standard error of calibration?SEC?was 0.539,and the standard error of cross validation?SECV?was 0.641 in the range of 550-1068 nm and 1108-2180 nm by using a 2,6,6,1 math treatment.The determination coefficient of prediction(RSQ_val)for near-infrared predictive protein values to chemical values of the prediction sample set reached0.9144.t-test of paired data indicated that there were no significant differences between reference values and prediction values of the model.The relative prediction deviation was greater than 3.Thus,the model is accurate,precise,and stable and can be applied in practice.?3?Near-infrared spectroscopy model testing oil content:the determination coefficient of cross validation?1-VR?was 0.905,the determination coefficient of calibration(RSQ_cal)was0.923,the standard error of calibration?SEC?was 0.515,and the standard error of cross validation?SECV?was 0.572 in the range of 888-1080,1100-1758 and 1968-2208 nm by using a 2,6,6,1 math treatment as well as a standard normal variant plus detrending?SNV+DET?scatter treatment.The determination coefficient of prediction(RSQ_val)for near-infrared predictive oil values to chemical values of the prediction sample set reached 0.8486.t-test of paired data indicated that there were no significant differences between reference oil values and prediction values of the model.The relative prediction deviation was less than 3 and greater than 2.The model is accurate and precise,and can be used for crude determination and screen of oil content.?4?Near-infrared spectroscopy model testing gossypol content:the determination coefficient of cross validation?1-VR?was 0.783,the determination coefficient of calibration(RSQ_cal)was 0.858,the standard error of calibration?SEC?was 0.034,and the standard error of cross validation?SECV?was 0.042 in the range of 2100-2498 nm by using a 4,10,10,1math treatment as well as an inverse multiple scattering correction?IMSC?scatter treatment.The determination coefficient of prediction(RSQ_val)for near-infrared predictive gossypol values to chemical values of the prediction sample set was 0.6414.The relative prediction deviation was less than 2.The model is still used for prediction.However,further improvement of the model is required.