Research Of Rapeseed Quality Detection Method Based On Micro Near Infrared Spectrometer

As a rapid,nondestructive and non-polluting analytical method,near infrared(NIR)spectroscopy is widely used in the production process of many industries,especially in the field of agricultural products.There are advantages of high resolution and good stability in NIR spectrometers for laboratory uses,but such spectrometers have disadvantages of slow scanning speed,high price,large volume and poor portability which has a certain influence on the realization of on-site rapid detection,and hinders the development of near-infrared spectroscopy technology for field detection.Micro-electro-mechanical system(MEMS)technology is a high-tech based on semiconductor micromachining technology.There are advantages of micro-volume,low cost and high sensitivity in Micro-NIR spectrometers based on MEMS technology,which can play an important role in field detection in the future.A miniature near infrared spectrometer called "N100" based on MEMS technology was developed by us independently.And the research for detection methods of rapeseed quality was launched with rapeseed crude fat and crude protein as the main research objects.Specifically,the following research work was conducted:1.To evaluate performance of N100,the indices of output signal,absorbance repeatability and baseline stability of N100 were tested and evaluated.The results show that in the optimum working point(output signal level in 20 000~40 000 Counts range),the absorbance repeatability is the best at 10 ms integration time,with the relative standard deviation(RSD)stabilized at around 0.003 and the RSD of 100%T stabilized at around 0.000 4.It shows that N100 basically fulfills the precision requirement in rapid detection of rapeseed quality indices during production process.2.To measure the effect of particle size of test materials on near-infrared spectra,flour,millet and rice which have different sizes were selected as samples,and stability of the original light signal intensity of samples were verified.The results show that the light signal intensity of flour was the most stable,whose RSD was stabilized at about 0.001 5,and that of rice was the most unstable,whose RSD was stabilized at about 0.004 5.It is indicated that the stability of the near infrared spectra is gradually decreasing with the increasing of particle size of test material.In order to ensure the stability of the spectra,it can be handled by some necessary physical methods such as crushing for larger granular materials.3.To further optimize the performance of N100,45 ° and 60 ° were designed as the incidence angles of light source,and spectral stability test was verified by the spectral output signal of rapeseeds collected under both angles.The results showed that the spectral output signal of materials was more stable under 60°,whose RSD was stabilized at about 0.003,but for 45°,RSD was stabilized at about 0.007.And the significance of spectral difference between the two angles was verified by the F test in mathematical statistics.So,for quality assessment of rapeseeds,the accessory piece with 60 ° incidence angle of light source should be used.4.To verify whether N100 could meet the requirements in actual testing,104 rapeseed samples from different regions were collected,and the rapeseed spectra were collected using N100 and FOSS 6500 spectrometers for determining their crude protein and crude fat contents.The correction models of crude fat and crude protein were created using data processing software called NIRS.The results showed that,for N100,the prediction result of crude fat PLSR correction model with Duplex + first derivative + abnormal samples emoved was the best.Its RC,RMSEC,RP and RMSEP are 0.8829,1.1544,0.8459 and 1.5908 respectively.The prediction result of the crude protein PLSR correction model with KS + MSC + abnormal samples removed was the best.Its RC,RMSEC,RP and RMSEP are 0.9181,0.4700,0.8469,and 0.5839 respectively.And for the prediction result of the crude fat PLSR correction model by FOSS 6500,RC,RMSEC,RP and RMSEP are 0.9440,0.8885,0.9267 and 1.1445 respectively,for that of the crude protein PLSR correction model,RC,RMSEC,RP and RMSEP are 0.9777,0.3202,0.9627 and 0.4015 respectively.The results showed that N100 basically achieves the requirement of quality detection of rapeseeds in real life.However,its performance was not as good as the relative models established using the high-precision desktop instrument “FOSS 6500”,suggesting that N100 has large room for improvement.5.To solve the problem of temperature factors influencing the robustness of correction model,the content of crude fat of rapeseeds was taken as research object.The PLSR correction models of crude fat were established by using spectral pretreatment and temperature mixing correction model respectively.The results showed that,after spectral pretreatment(e.g.by MSC),RMSEP values(1.9167 and 2.3410)of the verification set at 22 ℃ and 28 ℃ were close to RMSEP value(1.4934)of the verification set at 25 ℃.However,their RAD values(4.230 and 4.781)were significantly higher compared to RAD(2.651)of the validation set at 25 ℃.And after model establishment through temperature mixing correction,RMSEP values(1.4457 and 1.3904)and RAD values(3.113 and 2.887)at 22 ℃ and 28 ℃ were very close to RMSEP(1.3904)and RAD(2.521)of the validation set at 25 ℃.It showed that the simple spectral preprocessing method has no obvious effect on establishing a robust correction model,whilst the method of temperature mixing correction model is effective for establishing the robust correction model.