| In the world, China is the biggest country of fruit production, however, due to the limits of the postharvest technology of quality detection, commercialization of processing, the criteria of evaluating fruit quality, the exports volume of China’s fruits is very small, and it is hard to enter the international high-level market, which greatly influenced our country’s fruit competitiveness in the international trade and foreign exchange earning ability. Nondestructive detection of fruit quality by near infrared spectroscopy technology can realize on-line automatic classification of fruit, and improve their commercializations. The technology with the characteristics of real-time, nondestructive, convenience and high accuracy, has become one of the most rapid detection technologies since1990s. It has been widely used in food, medicine, agriculture and other areas. In this respect, our country starts to explor this technology later than foreign countries, but we had done a lot of researches by using near infrared technology and achieve some outstanding results.This research topic is mainly about the foundmental research of on-line detection of soluble solids content (SSC) in Crown Pear by visible/near infrared (VIS/NIR) Spectroscopy with comprehensive utilization of optical fiber sensing technology and chemometrics methods during the research of soluble solids (SSC) of pear. After setting up an on-line detection system, and an evaluation method of the stability of on-line detection system was put forward. The automatic and manual reference spectra collection models were studied, and different wavelengths selection methods and modeling methods combined with different spectral preprocessing methods were established for pear’s SSC online prediction. The optical properties of the reference with different thickness and different glass fiber content are studied.In this paper, the main research contents and research conclusions are as follows:(1) Research on the evaluation of the stability of on-line measurement system. After setting up a measurement system, how to evaluate whether this system satisfies the spectrum acquisition? This study proposed the signal-to-noise ratio (SNR) and spectral area change rate (ACR) to evaluate the stability of the VIS/NIR spectroscopy detection system. Standard reference spectra were collected under the different light sources (LS1and LS2). Spectral detection system under the condition of LS1obtained higher SNR and smaller spectral area change rate, and it was confirmed by traditional method. So we can get the conclusion that SNR and spectral area change rate is feasible to evaluate the stability of the measurement system.(2) Research on the reference spectrum collection modes. We compare the influences of the manual and automatic reference spectrum collection methods to model precision. The results show that automatic reference collection mode does improve the precision of the model in a certain extent; we also study whether it is necessary to collect the reference spectra for more times in the process of measurement. Under the condition of only one reference spectrum collection in the whole measurement, the prediction results of manual and automatic collection manner were rpre=0.8912, RMSEP=0.379°Brix and rpre=0.8987, RMSEP=0.359°Brix, respectively. Under the condition of one reference spectrum after every20fruits, the prediction results of manual and automatic collection manner were rpre=0.8948, RMSEP=0.370°Brix and rpre=0.9097, RMSEP=0.342°Brix, respectively. Automatic on-line collection reference mode can not only improve the automation of the whole spectral system, but also can improve the accuracy of prediction model. So we think the promotion of automatic reference collection methods in the application of on-line detection system has very important significance.(3) The optimization research of the on-line model by key wavelengths selection. In order to select the optimal variables for detection model, competitive adaptive reweighted sampling (CARS) key wavelength selection method is used. Five different modeling methods, competitive adaptive reweighted sampling partial least square regression (CARS-PLSR), stepwise multi linear regression (SMLR), competitive adaptive reweighted sampling multi linear regression (CARS-MLR), competitive adaptive reweighted sampling stepmise multi linear regression (CARS-SMLR) and Full-PLSR have been carried out for comparation study. CARS algorithm can effectively choose key wavelength, CARS-PLSR, CARS-MLR, and CARS-SMLR modeling methods can all get good prediction results. CARS-SMLR chose10key variables for modeling, which only accounted for2.03%of full spectrum modeling wavelengths, and get a good prediction result with correlation coefficient of prediction (rpre)0.812and root mean square error of prediction (RMSEP)0.451°Brix, which is suitable for on-line dection of SSC of pear. Key wavelength selection methods can improve the precision of the model and largely simplified the detection model.(4) Research on the optical properties of the reference. Sample absorption spectrum is determined by reference spectrum, so how to ensure the stability of the reference spectra is significant for the subsequent spectral processing and on-line detection model. This research is based on the Teflon reference material. The changing rule of the transmission spectra of Teflon plates under the condition of different thickness,1cm,2cm,3cm and with different proportions of glass fiber,10%,15%,20%,25%were compared. Under the condition of100w tungsten halogen light source, the Teflon plate with1cm thickness and15%glass fiber content is the optimal plate reference solution. In the study of different sizes of Teflon reference ball, we get that the size of the reference ball and transmission spectra intensity do not satisfy the linear change rule in all seven grade diameters,70mm reference ball got the maximum intensity of transmission spectra,65mm reference ball got the minimum transmission spectra. |
PDF Full Text Request