Research Of Influence Factors On Determination Of Fruit Quality By Visible And Near Infrared Spectroscopy

Nowadays, the total fruit output of China ranks the first place in the world. For the lack of effective technique to determine the internal quality of fruit, there are serious problems, such as a low taste and nutritional value with a good appearance, in the development of the Chinese fruit industry. Visible and near-infrared (Vis-NIR) spectroscopy technique, as a nondestructive detection method, is more and more widely used for the real-time monitoring and grading of fruit quality. However, it remains difficult to detect the actual quality of agricultural products with high accuracy due to the lack of control at every procedure of modeling. And the applicability of a model is limited to its modeling parameters and the physiological variance of samples. In this article, some overlooked but crucial influences would be analyzed in the process of analyzing, including the performance of chemometrics, the difference of instrument, the testing area of reference values, peel of fruit, the scanning location of spectra, and the interaction between properties.1. In order to find the appropriate pretreatments for determination of fruit quality, the interferences were simulated by mathematic method, and the calculated spectra were compared with ones of real samples. The baseline deflection (L2-L1=mx2+nx+k) and spectral deviation (L2=a·L1+b) of the ture spectra L1 were two main kinds of interference information in the apparent spectra L2· And the derivative and multiplicative scatter correction (MSC) were found to be useful to reduce the noise and to highlight the effective signal.2. Three portable Vis/NIR instruments, named (1) ACCUNIR2200, (2) K-BA100R and (3) SupNIR-1100, were given to test their performances of detection modules. And all the parameters, including the noise of baseline, the repeatability of baseline, the accuracy of wavelength, the repeatability of wavelength, and the repeatability of absorbance, had met the design requirements and featured reliable operation. After scanning spectra by these devices, three quantitative models were built to detect the SSC values of apple. Their results were (1) Rp=0.938, SEP=0.429; (2) Rp=0.945, SEP=0.323; and (3) Rp=0.918, SEP=0.478, which met the requirements of predicted accuracy (SEP<0.7 °Brix). The K-BA100R had the best performance because of its appropriate optical components and scanning mode of interactance. It was necessary to choose a suitable instrument according to its special construction and real performance of testing.3. Three testing areas of apple for determination of SSC were (1) the outer flesh cuboid (Part C, 2.5×2.5×1.5 cm3), (2) half of peeled apple, and (3) half of intact apple. According to the analysis of variance (ANOVA) method, sampling methods of SSC had significant effect on quantitative accuracy, and the model, using SSC values detected from Part C, provided the best results. The light-passed area was helpful to improve the relationship between the chemical values and the corresponding spectral matrix.4. The original spectra were scanned both with unpeeled and peeled samples. In the region 500～1010 nm, fruit peel had significant effect on the signal in the visible wave band. However,730～932 nm could be used to weak the influence of peel on the accuracy of models. In addition, the evaluation ability of this technique could be improve again when scanning the spectra of samples after peeling. On the basis of these two ways, the quantitative model of SSC was built, and the results were Rp=0.939, RMSEP=0.282.5. A statistical analysis was performed on a large set of spectral data (500～1010 nm) to analyze the influence of testing location (the stylar end and the equator area) of mini-watermelons on its soluble solids content (SSC). Period of growth and season of harvest were responsible for most of the biological and spectral variability in fruit from different geographical locations. With respect to these two factors, the robustness of the calibration models for SSC was evaluated based on internal and external validation. Results showed that the accuracy of our model increased when we applied the spectral and chemical information from the stylar end (750～950 nm), because of its high sensitivity to stage of maturity (Rp=0.913, RMSEP=0.473). However, it is desirable to establish a global model using data (730～932 nm) from the equator-area for samples of different maturity and collected in different seasons, which may improve calibration for future measurements. There was a difference in the spectra from the varied positions of the same fruit, which directly affecting the reliability of modeling data. According to the actual situation of individual fruit, the acquisition location should be selected to meet the specific requirements of application.6. In the development of mini-watermelon, SSC, lycopene, and moisture are three important indexes to evaluate maturity of fruit. The interactions between each index lead to the overlap of their absorption peaks. Based on both correlation coefficients and regression coefficients between SSC values and the spectral data transformed by second Savitzky-Golay derivative, effective modeling wavelengths were selected, including 744,826,854,886, and 912 nm for SSC; 744,786,826,886, and 912 nm for moisture; 548,700,806,826,886, and 912 nm for lycopene. The accuracy of MLR models using these variables was close to that of PLS models with 750-950 nm. In order to improve the predicted ability for watermelon quality during the growth period, multivariate multi-block technique Factor Analysis enabled integration of all these traits to establish a comprehensive indicator. And this indicator was assessed noninvasively to evaluate the maturity of watermelon with high accuracy and visualization.