Optimization Of Rapid Analysis Model For Oil Content And Fatty Acids In Rapeseed Using Machine Learning

As an important oil and cash crop in China,rapeseed has always occupied a large proportion in the supply of plant edible oil and national economy.Traditional chemical methods for quality detection of rapeseed without exception need to use chemical reagents,complicated operation,generally long cycle,high cost.Near infrared reflectance spectroscopy(NIRS)analysis with nondestructive,fast,low cost of a variety of advantages,very suitable for mass screening of quality traits of crop breeding,the near infrared spectroscopy in determination of oil plays an important role in the specific quality,chemical methods,closely connected to the near infrared spectrum technology is applied to measuring the oil product quality.The increasing economic level of Chinese people makes their demand for edible vegetable oil more and more high,the requirement is also more and more high,how to more efficient discrimination of rape parameters is very important.NIR spectroscopy is used to obtain data and establish a more accurate oil content model of rapeseed,which can greatly improve the detection efficiency and facilitate the subsequent comprehensive evaluation of rapeseed germplasm.To provide a guarantee for the early stage of rapeseed breeding.This paper takes brassica rapeseed as the research object,the main work is as follows:(1)1,400 samples of experimental rapeseed were selected for near infrared spectroscopy processing analysis,and 200 samples were analyzed by gas chromatography to measure fatty acid data,residual method to analyze and measure oil content data,and seed image and weighing processing to measure related attributes.(2)using XGBoost algorithm using near infrared spectra of raw data to establish forecast model of oil content of the RMSE is 0.58,the chemical method is used to replace the original data to establish forecast model of oil content of the RMSE is 0.3,the results showed that oil content of the prediction model is established to replace data is better than using the raw data to establish predictive model of the oil content.The optimization effect has been achieved.Three machine learning methods,SVR,XGBoost and BPNN,were combined with the data replaced by chemical methods to establish the oil content prediction model of rapeseed based on near infrared spectroscopy.RMSE were 1.31,0.58 and 10.94,respectively.The model built by the XGBoost algorithm works best.(3)The prediction models of erucic acid,oleic acid,linoleic acid and linolenic acid based on NIR spectrum were established by XGBoost algorithm combined with the original data of NIR spectrometer,with RMSE of 3.25,3.68,3.13 and 4.2,respectively.The fatty acid data obtained by gas chromatography were replaced.The RMSE values of erucic acid,oleic acid,linoleic acid and linolenic acid were 3.81,3.67,3.63 and 3.64,respectively.The prediction model of erucic acid and linoleic acid using the original near-infrared spectral data is more effective.It is better to use the substituted data to establish the prediction model for oleic acid and linolenic acid.(4)the correlation analysis to find the images and weighing rapid analysis to identify seeds and seed oil content in the parameters of the correlation coefficient is larger,combined with the oil content of the data obtained from residual method,respectively with the SVR,XGBoost,BPNN three machine learning method based on image and weighing rapid analysis to identify seeds of rapeseed oil content prediction model.RMSE were 1.47,2.19 and 10.46,respectively.The SVR model is the best.