Detection Methods For Processing Tomato Soluble Solids Content Based On High Spectral Technology

As the major commercial crop in Xinjiang province, processing tomato has become one of the localexport-oriented industries for its wide cultivating and high productivity. With the increasing demand ofinternational market for food quality, manual sorting cannot meet on-line testing of processing tomatoquality. By considering Li Ge87-5and Jin Fan No.1widely cultivated in Xinjiang as test samples, thispaper focuses on nondestructive test technique for soluble solids content by using hyperspectral imagingtechnology and chemometrics method, which was the important indicator of processing tomato quality. Themain conclusion of the thesis is as follows:(1) For the difference of individual size, a highly adaptive fixed device was fabricated such that theobject distance consistency of sample top and camera can be ensured in hyperspectral image acquisitionprocess.(2) The band551.07nm with great spectral difference between the regions of interest and backgroundwas selected, and then the batch processing of the extraction for the region of interest was achieved basedon band threshold value method, and the average spectra was also computed in MATLAB programmingmethod.(3) By modeling and analyzing spectral information of different spectral profiles (The equator circularsurface, Areola surface, Pedicel surface) for Li Ge87-5and Jin Fan No.1, the better detection surfaces fortwo test sample were obtained. The results indicate that equator circular surface was adopted for Li Ge87-5and Jin Fan No.1.(4) The impact of model precision among different modeling parameters and preprocessing method iscarried out.1）Three methods of baseline correction with constant, MSC and SNV were discussed. By comparison,MSC baseline correction was employed for Li Ge87-5and Jin Fan No.1.2）The modeling regions were selected by using concentration coefficients, and then the band of560~930nm was determined as an optimal modeling region.3）Smooth results of the first derivative spectra and the second derivative spectra with Savitzky-Golayconvolution and Norris differential are compared, and then Norris differential smoothing was determined.(5) By analyzing the predictive effect of two test samples based on PLS, PCR and SMLR, an idealmodeling approach was obtained. PLS has better performance for modeling second derivative spectrum ofLi Ge87-5, where correlation coefficients of calibration set and prediction set equal to0.790and0.835,respectively, and the root mean square errors of calibration set and prediction set were about0.292%and0.319%, respectively. PLS gived better results on modeling original spectrum processed by MSC for JinFan No.1, where correlation coefficients of calibration set and prediction set equal to0.730and0.771,respectively, and the root mean square errors of calibration set and prediction set were0.269%and0.316%,respectively.(6) The prediction module of soluble solids content of processing tomato was established based onMATLAB. By using this module, the sample spectra can be imported, and the extraction and preprocessingof hyperspectral image for the region of interest were achieved. Moreover, the selections of calibration setsample and modeling region were completed, and the modeling by using PLS and BP neural network werealso carried out. Consequently, our work provides some convenience for further research.