| China is the largest apple producing and consuming country in the world. Apple plays a critical role in China’s modern agricultural economic. With the international trade competition and apple consumption pay more attention from appearance quality to internal quality and safety, currently, postharvest commercial processing technologies for apple internal quality detection and safety analysis are urgently needed especially the rapid non-destructive, online detection technology and systems. The objective of this study was to investigate the non-destructive and fast quality and safety analysis and assessment method for apple products. The project developed rapid, high precision, non-destructive, online, intelligent detection technology and system, and will be quite helpful to improve the quality of apple, reduce the apple postharvest economic loss, and enhance market competitiveness. In this study hyperspectral imaging, near infrared spectroscopy and information fusion technology were used to predict quality and safety of apple. From the three aspects of detection mechanism, key technology and methods and system, the main contents and results of the research are as follows:1. Light propagation properties of apple tissues were investigated to explore the effective depth of penetration and scattering distribution. First, optical transmission testing platform was designed to study the characteristics of light attenuation and rule of light distribution. The effective penetration depth of apple tissue was calculated by least squares fitting method. Meanwhile, the influence of apple peel for optical penetration depths was analyzed. The method of nonlinear curve fitting was used to explain the spatial resolution attenuation law, and to characterize the optical transmission scattering properties. The results of texture properties were objective and accurate by the quantitative description of instrument. Texture determination method was established to perform the texture profile analysis. The differences of texture in different area of apple was compared and analyzed. The overall results of this study will be helpful to provide theoretical basis and system design for nondestructive detection of apple quality.2. Hyperspectral imaging integrating both spectroscopic and imaging techniques with higher spatial and spectral resolution has been developed to study the quality of fruit. The influences of modeling accuracy affected by shape and size of ROI were analyzed. The optimum results were achieved when roundness ROI with diameter of150pixels was extracted. The correlation coefficient (Rc) was0.9305in the calibration set; and the Rp was0.9232in the prediction set, respectively. This study proposes a method for correcting the light intensity of the radiation non-uniform for spherical objects. The sugar content distribution map was generated based on the spatial positions of pixels. Then, the visualization of sugar content distribution in apple was achieved by pseudo-color mapping. Shortwave near infrared (SWNIR) and long wave near infrared (LWNIR) spectroscopy associated with synergy interval partial least squares (siPLS) method was investigated and compared for predicting quality attributes of apple. The optimal models were achieved with#,,=0.9228for soluble solid content (SSC),Rp=0.7543for peel hardness,Rp=0.8543for flesh mean firmness, and Rp=0.8487for active acid, respectively.3. Ant colony optimization algorithm combined with partial least-squares (ACO-PLS) was employed to select the characteristic wavelength of NIR spectroscopy for apple SSC prediction from different geographical region. By using the features of heuristic global search and Monte Carlo roulette random selection mechanism; ACO explored optimally the efficient wavelength from the NIR spectroscopy of apple to develop models for predicting the SSC of apple. Good prediction performance was obtained for SSC with Rp=0.9708, and RMSEP=0.5144, respectively. Experimental results showed that the performance of ACO-PLS model was superior to the performances from traditional PLS and genetic algorithm (GA) models with the least variables. The study demonstrates that ant colony optimization could effectively select the characteristic wavelengths of NIR spectral to improve the model robustness and applicability.4. In order to control the accuracy of the calibration models for SSC with respect to color affects, color compensation method for SWNIR and LWNIR spectroscopic techniques to measurement SSC of apple fruit is proposed. Multivariate calibration techniques were compared with linear and nonlinear regression methods, including feature and latent variable construct techniques. Color space coordinates have been integrated in order to compensate for color variation in NIR spectra. Color compensation models were also established and compared. Independent component analysis-support vector machine (ICA-SVM) compensation models obtained excellent performance both SWNIR and LWNIR. The models produced prediction accuracy with Rp=0.9398for SWNIR while Rp=0.9455for LWNIR, respectively. The results revealed that color compensation could significant improve the prediction accuracy for SWNIR. Considering the commercially viable application, SWNIR techniques coupled with color compensation can be effectively applied in the industry as an analytical tool to monitor the quality of apple.5. Shortwave near-infrared spectroscopy for on-line detection system was developed to determine internal quality of apple. The critical point in application for on-line implementation of NIR spectroscopy is to build good robustness and high accuracy quantitative analysis models. In order to obtain reliable spectra, intensity normalization correction was performed. GA, successive projection algorithm (SPA), and ACO were studied to select feature variables, and combined with PLS to determine the sugar content of apple, respectively. The prediction model built using ACO-PLS was the best with Rp=0.9358, and RMSEP=0.2619°Brix, respectively. The study demonstrated that NIR combined with variable selection methods was feasible to improve on-line determination models of sugar content in apple. Furthermore, the successful scale-up of the method proved its capability to be implemented in the manufacturing plant.6. Simultaneous determination internal quality and hidden defect in apple using online near-infrared NIR transmittance spectroscopy was proposed. Near infrared diffuse reflectance spectroscopy could not fully and deep capture the internal quality information due to limited light penetration depth. In view of the difficulty of apple internal quality and hidden defects detection, an online determination system based on NIR. transmittance spectroscopy was developed and used for signal acquisition related to the internal quality and hidden defects. Combination of system parameters will be optimizing to improve the quality of online transmission spectra. The stepwise multiple linear regression (SMLR) models were developed to relate absorbance spectra of apple samples and SSC. Through the independent sample verification, the correlation coefficient of prediction was0.8764. Pattern recognition algorithms for classification of defect samples with linear discriminant analysis (LDA) yielded90.14%overall classification accuracy in the validation sets.This research provides new ideas for rapid nondestructive quality detection of apple. The results are promising for the promotion of the role of nondestructive detection technology for practical, which is also of great significant in improve the level of detection in the fruit industry in our country. |
PDF Full Text Request