| It is the hot research field of fruit quality automation measurement and classification in international agricultural engineering and food engineering technology. This research will satisfied the consumer demand in food quality and food safety, improve the supermarket value and competition of fruit production and increase the farmer income. China is the largest country of the total yield of fruit in the world. But the output yield is only 3.0 percent of total yield for its low process technique which is not consistent with the first country of fruit total yield. With the improvement of people life quality and abundance of varieties fruits, the higher demand quality of fruit was needed by consumer at present. The quality of fruits was mainly determined by sugar content and organic acid which it not only affects the taste and color of fruit, but also acts as a prerequisite for synthesis of fruit vitamin and carotenoids. Routine analytical methods used for fruit internal quality measurement are destructive, expensive, complex, time and labor consume and off-line by nature. It is very urgent to find mechanism and mathematic models used to nondestructively measure fruit internal quality. The research will expend the national high technology of "863" program and combine some other cross knowledge in order to find the mechanism and mathematic algorithms of measuring fruit sugar content and acidity. The mathematic models for nondestructive measurement sugar content and acidity of apple and pears were also developed. It is an important advance to fruit quality measurement technology and the study will lead to solve the problem of the long time of sample preparation, subsequence processing difficulties and so on.In the dissertation, apples and pears are the main study objective and the sugar content and acidity indices about apple and pear are the main testing indices. The nondestructive measurement mechanism and methods using NIR, testing optimization parameters, influence factors of mathematic calibration models, linear and nonlinear mathematic models, and the effective signal extraction were investigated using orthogonal experiment design. The main novelty and creations of the dissertation involve:1. The modern instrument and experiment analysis techniques were applied to experiment research and the measurement factors including instrument status, sample parameters, sampling and testing conditions were analyzed and the effects on the nondestructive measurement prediction to fruit sugar content and acidity were also investigated. An important testing parameters about fruit sugars and acidity measurement were given. It will be given an important reference value to explain the best testing position, best testing distance and sample status and give the same testing standards. The main absorbance positions of fruit are found at the wavenumber of 10176 cm, 8313 cm-1, 6860 cm-1, 5172 cm-1 and 4321 cm-1 and there is no obvious different with different samples. By the orthogonal experiment design, the best matching parameters used to FT-NIR measurement are as the following: scan number is 64, resolution is 16 cm-1 and plus gain is 4. It would be concluded that the calibration models used the best matching parameters will given the best prediction accuracy.2. The nondestructive measurement used to fruit sugar content and acidity were studied, and the linear and nonlinear mathematic models were developed. The linear mathematic models for sugar content (SC), titrated acidity (TA) and valid acidity (pH) of apples and pears were developed using multiple calibration regression methods. Different spectral pretreatment and different selected spectralrange was used to optimize the calibration models. The best calibration models of SC, TA and pH of apple samples at the wavenumber of 4242.63cm'1-12303.62 cm"1, give the rcof 0.954, 0.790 and 0.821, the RMSEC of 0.540, 0.041 and 0.069, the RMSEP of 0.530, 0.043 and 0.066, the rcmss of 0.947, 0.750 and 0.720, respectively, when the factor numbers used to PLS are 9, 10 and 10, respectively. The best calibration models of SC, TA and pH in pear samples at the wavenumber of 4504.90cm"'-7112.19 cm"1, give the rc of 0.923,0.85 and 0.88, the RMSEC of 0.505, 0.22 and 0.132, the RMSEP of 0.544, 0.039 and 0.083, the rcrossof 0.911, 0.840 and 0.870, respectively, when the factor numbers used to PLS are 10, 8 and 10, respectively. These mathematical models will be applied to fruit quality sorting and processing industry. The nonlinear mathematic algorithms by the combination of the wavelet transform and the matrix BP artificial neural network method were investigated. The best calibration models of SC,, TA and pH in apple samples give the rcof 0.950, 0.941 and 0.925, the RMSEC of 0.199, 0.013 and 0.100, the RMSEP of 0.217, 0.023 and 0.200, the rcmss of 0.955, 0.845, 0.857 and 0.970, respectively when the input nerve cell number is 6, nerve cell number of hidden nodes is 4, the momentum is 0.2 and learn ratio is 0.5 with the grades descending algorithms.3. Because the long calculating time using cross-validation of all samples method, the optimization method of fruit sample based on Genetic Algorithm (GA) algorithms was investigated and the mathematical model using (rcross/1+RMSECV) as object function was established. Outlier and similar sample diagnosis and eliminating algorithms were also investigated. It can be concluded that one sample residual is more three times than the average residual of all samples, the sample will be eliminated as an outlier. The chromosome coding and decoding methods for parameter and the disposing method for restriction condition were studied. The influences on the results using cross-probability, variance-probability, evolution algebra, and the pop-size were analyzed. The best running and matching parameters of GA are as the following: Popsize is 30, Maxgen is 60, cross-probability is 0.6 and variance-probability is 0.01. The precision of the mathematical model optimized by GA is obviously improved with the high-correlation coefficient of prediction model and the low prediction standard error.4. The mechanisms of simultaneity quantification of the glucose, fructose, sucrose, total sugar and reducing sugar content were investigated and the calibration models used to non-destructive measurement of these constituent sugars were developed by partial least square (PLS) regression. The best calibration models of frucrose, glucose and sucrose of apple samples with the 2" deviated spectra and the wavenumber of 4296.62cm"'-9090.90 cm"1, give the rcross of 0.955, 0.845, 0.857 and 0.970, and the RMSEP of 0.313,0.354,0.446 and 0.507, respectively, when the factor number used to PLS are 8, 7 and 4 respectively. The best calibration model of reducing sugar in apple samples with the 1st deviated spectra and wavenumber of 5453.70cm"'-12103.06 cm"1, gives the rc of 0.923, the RMSEC of 0.271, the RMSEP of 0.485, the rcmss of 0.76, respectively, when the factor number used to PLS is 4. These studies will help to solve the problem of the complicative chemical analysis, high cost, long testing time, low accuracy and so on. |
PDF Full Text Request