Analysis Of Sunflower Seed Quality Based On Near Infrared Spectroscopy

In the past 40 years,sunflower seed production has developed very fast that its output ranks second among oil crops,second only to soybeans.The quality of different batches and different quality sunflower seeds is different,so the price is also very different.At present,there are many varieties of sunflower seeds on the seed market,which are mixed with fish and dragons.Frequent occurrences of sub-filling have greatly damaged the interests of farmers..In order to ensure the consumer's rights and effectively identify the oil quality of sunflower seeds,it is urgent to establish an effective detection method to quantify main content of sunflower seeds,such as protein,water and oil and qualify the sample.for identification.In recent years,near-infrared spectroscopy has been widely used in various agricultural testing fields due to its fast-online analysis capability and its non-destructive and non-polluted characteristics.In this paper,the sunflower seeds of the oil census crops are provided by Shanghai Philippine Optics Co.,Ltd.The protein,oil and water content are determined by standard chemical methods.The near-infrared spectrum of the sample is determined by a spectrometer jointly developed by the laboratory and the company.And we use metrological method to analyze and quantify the near-infrared spectrum of sunflower seeds.In this paper,several key problems in the identification and quantitative detection of near-infrared spectroscopy are discussed in depth from the aspects of spectral pretreatment,wavelength screening method and calibration model establishment.Near-infrared is carried out to correct the prediction accuracy of the model.Spectral discrimination of different sunflower seed quality studies.Designed to provide a new reference for non-destructive testing of sunflower seed quality.The main research work is as follows: 1)This paper compares the effects of various pretreatment methods and method combinations on the spectral data of sunflower seeds in the same band,and finds the optimal corresponding three indexes of protein,water and oil in used several pretreatment methods.The effects of the three-band optimization method,such as the irrelevant information vector elimination method,on the partial least squares(PLS) model are compared under the premise of not affecting the accuracy of the analysis.The research shows that the wavelet transform has the best preprocessing effect,and the band selection effect under the irrelevant vector elimination is optimal.2)Using the orthogonalization-based partial least squares dimension reduction method to classify and identify the spectral samples.The simulation results show that the accuracy of classification identification is as high as 92.30% and the lowest is 84.21%.The near- infrared spectroscopy method using partial least squares(PLS)and BP neural network was used to quantitatively model the sunflower seed spectrum,and the model was verified by the internal cross-validation method of the leave-one method to test the root mean square error of the sample.And the percentage error is the evaluation index.The results show that BP neural network is the most ideal in the quantitative model established by two different chemometric methods.3)Based on the orthogonalization-based partial least squares dimension reduction method,the sunflower seed spectral data and protein,water and fat content were established under several ideal pretreatment and band screening methods.Between BP neural network and PLS-based quantitative analysis model.The results show that the quantitative analysis of the three components related to the quality of sunflower seed samples is the most accurate after classification by PLS based on the orthogonalization-based partial least squares dimensionality reduction method.The effect is better than the unclassified sunflower seed BP neural network model.The classified BP neural network model may be the worst of the four models because of the low generalization ability of the sample.