Study on Rapid Non-Destructive Inspection of Apple’s Quality by NIR Spectroscopy and Hyperspectral Imaging Technique

论文摘要

Near-infrared （NIR） spectroscopy and hyperspectral imaging are being used in various applications in agricultural and food quality detection, and evaluation. The present study attempted to the feasibility on detecting apple quality by NIR spectroscopy and hyperspectral. Some questions in apple quality detection were emphasized on. NIR spectroscopy and hyperspectral imaging in this research were used to determine internal qualities; mainly sugar content and fruit firmness, and external quality; bruises of apples. This study hoped to improve the level of apple quality detection in procedure of apple process, distribution and sale. Therefore, it is able to produce high efficiency, excellent quality and safety apple production. The main points are as follows:NIR spectroscopy was used for estimating sugar content and fruit firmness of apples. Spectral information was collected on the surface of the fruit by FT-NIR spectrometer in our laboratory. The FT-NIR spectrometer has a spectral range of 10,000-4,000 cm-1 （1,000-2,500 nm）, destructive analysis was carried out to measure the internal quality parameters and calibration models were established. The spectral data was preprocessed by Standard Normal Variate （SNV）, Mean Centering （MC）, Multiplicative Scatter Correction （MSC） and Min/Max normalization method and Partial Least Square （PLS） regression. In this study, the main points are as follow:The feasibility of the study was attempted to identify internal quality of apple simultaneously with NIR spectroscopy. In this experiment, the application of FT-NIR to predict quantitatively sugar content and fruit firmness of apples was established leading to a reliable calibration models and the number of Partial Least Square （PLS） factors. The applicability of multivariate calibration to NIR data has developed and proved. In order to build the model, first, the spectral region was selected, and the effects on the spectral preprocessing methods and number of PLS factors to results were discussed. The experiment results sufficiently performed the possibility of this non-destructive technique for measuring sugar content and fruit firmness in apples using spectral range of 10,000-4,000 cm-1 （1,000-2,500 nm） The PLS method had the potential to estimated the calibration and prediction model. It was found that the PLS calibration model in sugar content and fruit firmness determination were 9 and 12 PLS factors under Standard Normal Variate （SNV） and Multiplicative Scatter Correction （MSC） preprocessing method, respectively. The correlation coefficient between the NIR spectroscopy prediction results and reference measurement results could be reliable at the regression levels of R = 0.94106 for sugar content and R = 0.74893 for fruit firmness.The next study of NIR spectroscopy application attempted to determine sugar content and fruit firmness of apples using the different PLS algorithms. This research has investigated and compared the results provided by PLS, iPLS, and siPLS procedures for NIR quantitative analysis of sugar content and fruit firmness of apples. Various steps have to be executed in a calibration model. Compared among the PLS, iPLS and siPLS algorithm, the performance of siPLS model was the best. The optimal calibration model was achieved with R = 0.9250 and RMSECV=0.660 in calibration set and R = 0.9015 and RMSEP = 0.6899 in prediction set for sugar content and R = 0.8105 and RMSECV = 87.00 in calibration set and R = 0.6172 and RMSEP = 116.0529 in prediction set for fruit firmness. This study demonstrated that NIR spectroscopy with siPLS algorithm could be applied to determine the sugar content and fruit firmness in apples, and siPLS revealed its superiority in contrast with other multivariate calibration methods.For Hyperspectral imaging technique, as mentioned above, was highly advantageous in term of its data which can be presenting the information in both spatial and spectral information. In these researches, hyperspectral imaging was used to detect external quality （bruised defection） and internal quality （sugar content）. Hyperspectral imaging system has developed in our laboratory, which was arranged with spatial dimension of 1,280×600 pixels and with 1,024 spectral in the band of 408-1,117 nm. In this study, hyperspectral imaging was used for detecting both external and internal qualities. With following main points of study:This study first offered a new idea that apple external quality was discriminated with a new non-destructive detection technology, i.e., hyperspectral imaging technology. In this experiment, the potential of using a hyperspectral imaging technique was investigated for detecting bruises on "Fuji" apples. For this purpose, a hyperspectral imaging system has been built in the spectral region between 408 to 1,117 nm. Principal component analysis （PCA） was applied to determine the optimum wavelength of hyperspectral imaging. The top four PCs imaging were compared to select the optimal PC imaging which can most efficiently represent features of study target, then, the optimal wavelength imaging was determined according to the optimal PC imaging. An image processing algorithm for the optimal wavelength imaging was developed to detect the bruised regions of apple. The experimental results showed that the spectral region between 500-900 nm was most appropriated for detecting apple bruises. The 547 nm hyperspectral imaging was determined to detect apple’s bruise as the optimum wavelength imaging through PCA. Image processing based on morphological operation was developed to segment the apple’s bruise regions. The experimental results were obtained with the correct rate is 88.75% for bruise apples and 76% for nonbruise apples. This work indicated that it is feasible to detect bruise on apple with hyperspectral.The next study of hyperspectral imaging application was conducted to detect the internal quality of apple. The application of hyperspectral had three aims; firstly, to evaluate the internal quality attributes: i.e., sugar content and fruit firmness, based on the spectral information collected using hyperspectral imaging system, secondly, to select the optimal wavelengths for the determination of sugar content and fruit firmness, and thirdly, to develop the prediction models for predicting these sugar content parameters of apple from the spectral information. Results from the study have shown the determination by hyperspectral imaging technique using the wavelength range of 408-1,117 nm for estimating sugar content and fruit firmness. The PLS method has the potential to estimate the calibration and prediction model from their spectra. The PLS calibration model in sugar content and fruit firmness determination were 4 PLS and 5 PLS factors under SNV preprocessing method. The correlation coefficient between the hyperspectral imaging prediction results and reference measurement results were R = 0.90749 for sugar content and R = 0.60491 for fruit firmness. The PLS algorithm produced the calibration models which were developed to predict with reasonably good results and high correlation for estimating the sugar content and fruit firmness parameter of apple. It could be concluded that hyperspectral imaging technique is a potential useful for assessing internal qualities: i.e., sugar content and fruit firmness of apple.The overall results from this dissertation research demonstrated possibility for successful adoption of NIR spectroscopy and hyperspectral imaging technique for non-destructive analysis of apple’s external and internal quality parameters. Based on the results of the study conducted, it can be concluded that NIR spectroscopy and hyperspectral imaging technique have a high potential to detect other agricultural product quality in a nondestructive way. A reliable overall characterization of agricultural product quality may be obtained at a low cost. In comparison to other methods （including manual detection and chemical analysis）, these methods methioned in this work present some considerable improvement in detecting the quality of agricultural product.

论文目录

• ACKNOWLEDGEMENTS
• Abstract
• Table of contents
• List of Tables
• listof Figures
• CHAPTER 1 Introduction
• CHAPTER 2 OBJECTIVES
• CHAPTER 3 REVIEW OF LITERATURE
• 3.1 Assuring food safety and quality
• 3.2 Quality factors
• 3.2.1 Appearance （visual） quality factor
• 3.2.2 Textural （feel） quality factor
• 3.2.3 Flavour （eating） quality factor
• 3.3 Apple （Malus Domestica Borkh） and its quality
• 3.3.1 Quality properties of apple
• 3.3.2 Affecting factors on quality of apple
• 3.3.3 Apple's bruises, firmness and sugar content
• 3.3.3.1 Bruises
• 3.3.3.2 Firmness
• 3.3.3.3 Sugar content
• 3.4 Making the grade and apple grading
• 3.5 Non-destructive detection technique to certify safety and quality of fresh products
• 3.6 Machine vision system
• 3.6.1 Overview of machine vision system
• 3.6.2 Contributions of machine vision in this research
• 3.7 Near infrared spectroscopy technique
• 3.7.1 Overview of near infrared spectroscopy
• 3.7.2 Principle of near infrared spectroscopy
• 3.7.3 Application of near infrared technology to quality evaluation of fruits and vegetables
• 3.8 Hyperspectral imaging techniques
• 3.8.1 Overview of hyperspectral imaging technique
• 3.8.2 Principle of hyperspectral imaging technique
• 3.8.3 Components of hyperspectral imaging system
• 3.8.4 Advantages and disadvantages
• 3.8.5 Optimal band selection and combinations
• 3.8.6 Application of hyperspectral imaging to food quality & safety
• 3.8.7 The limitations of hyperspectral imaging technique
• 3.9 Quality evaluation of fruits and vegetables using machine vision, NIR spectroscopy and hyperspectral imaging technique
• 3.9.1 Machine vision
• 3.9.2 NIR spectroscopy
• 3.9.3 Hyperspectral imaging technique
• 3.10 Multivariate statistical methods for imaging and NIRS analysis
• 3.10.1 Typical of multivariate calibration methods
• 3.10.1.1 Principle component analysis （PCA）
• 3.10.1.2 Partial least square （PLS）
• 3.10.2 Wavelength selection
• 3.10.3 Pattern recognition classification
• REFERENCE
• CHAPTER 4 APPLICATION FT-NIR SPECTROSCOPY TO PREDICT SIMULTANEOUSLY SUGAR CONTENT AND FIRMNESS OF "FUJI" APPLES
• 4.1 Introduction
• 4.2 Materials and methods
• 4.2.1 Apple fruit samples
• 4.2.2 Spectral data acquisition
• 4.2.3 Reference measurements
• 4.2.4 Spectral preprocessing methods
• 4.2.5 Software
• 4.3 Results and discussions
• 4.3.1 Quantitative analysis of the PLS model
• 4.3.2 Sugar content
• 4.3.3 Fruit firmness
• 4.4 Conclusion
• References
• CHAPTER 5 DETERMINATION OF SUGAR CONTENT AND FRUIT FIRMNESS IN "FUJI" APPLES USING FT-NIR SPECTROSCOPY & DIFFERENT PLS ALGORITHMS
• 5.1 Introduction
• 5.2 Materials and methods
• 5.2.1 Sample preparation
• 5.2.2 Spectra collection
• 5.2.3 Destructive detection methods
• 5.2.4 Spectral preprocessing
• 5.2.5 Software
• 5.3 Results and discussions
• 5.3.1 Calibration of models
• 5.3.2 Sugar content
• 5.3.2.1 Results of PLS model
• 5.3.2.2 Results of iPLS model
• 5.3.2.3 Results of siPLS model
• 5.3.2.4 Discussion: The results of sugar content
• 5.3.3 Fruit firmness
• 5.3.3.1 Results of PLS model
• 5.3.3.2 Results of iPLS model
• 5.3.3.3 Results of siPLS model
• 5.3.3.4 Discussion: The results of fruit firmness
• 5.4 Conclusions
• References
• CHAPTER 6 APPLICATION OF HYPERSPECTRAL IMAGING TO DETECT BRUISES ON "FUJI" APPLES
• 6.1 Introduction
• 6.2 Principal of hyperspectral imaging
• 6.3 Materials and methods
• 6.3.1 Sample preparation
• 6.3.2 Hyperspectral image acquisition system
• 6.3.3 Image capture and calibration process
• 6.4 Results and discussions
• 6.4.1 Principal component analysis （PCA）
• 6.4.2 Selection of optimal wavebands
• 6.4.3 Asymmetric second different （ASD） and feature extraction
• 6.4.4 Bruise detection results
• 6.5 Conclusion
• Reference
• CHAPTER 7 NONDESTRUCTIVE MEASUREMENT OF SUGAR CONTENT AND FRUIT FIRMNESS FOR APPLES USING HYPERSPECTRAL IMAGING TECHNIQUE
• 7.1 Introduction
• 7.2 Materials and methods
• 7.2.1 Sample preparation
• 7.2.2 Hyperspectral imaging system and data acquisition
• 7.2.3 Hyperspectral imaging data processing
• 7.2.4 Reference measurements
• 7.2.5 Spectral preprocessing analysis
• 7.2.6 Software
• 7.3 Results and discussions
• 7.3.1 Quantitative analysis of the PLS models
• 7.3.2 Sugar content
• 7.3.3 Fruit firmness
• 7.4 Conclusion
• References
• CHAPTER 8 GENERAL CONCLUSION
• 8.1 NIR spectroscopy
• 8.2 Hyperspectral imaging technique
• RESEARCH PUBLICATIONS DURING THE Ph.D.
• AWARDS WON DURING THE Ph.D.
• APPENDIX
• Appendix A Table of sugar content of apple fruit sample in the year 2007
• Appendix B Table of fruit firmness of apple fruit sample in the year 2007