Detection Of Berry Quality Based On Near Infrared Spectrum And Machine Vision Technology

The trend of cultivation and production for berry is rapidly increasing in China. The berry's products have high requirement in internaltional markets. Due to the lack of uniform detection standard and technology, and the detection and grade for berry are carried out by using artificial method, as well as the speed of detection is slow, the competition of berry's products is poor in international market. Therefore, it is necessary to to evaluate the berry quality using the detection of near infrared spectroscopy and machine vision method, which have the practical significance and theoretical value to promote the economic benefit of berry industry by.In this research, the rapid detection for the nutritional quality was implemented by using near infrared analysis technology. On this basis, the variation of nutritional quality during the different harvesting periods was explored for the different varieties of blackcurrant, and the optimum harvesting period of Heifeng, Bajila, Yard, 94-4-13, Daisha and Browder were determined. 16 characteristic parameters of appearance for blackcurrant fruit was extracted by using machine vision technology, and BP neural network system for recognising variety of blackcurrant was constructed based on taking these quality characteristics of appearance as input variables. On the basis of entropy weight coefficient method, the blackcurrant quliaty was evaluated according to the area, long axis, average of minor axis, roundness, brightness, saturation, sugar acid ratio, VC content, anthocyanin content, and the grade standard of comprehensive quality for 6 varieties of blackcurrant was developed. The conclusions were obtained in this research as follow:1. According to single factor analysis of variance and orthogonal experimental design method, the influence of measuring temperature, scan number, variety, size and surface color on near infrared spectral response of blackcurrant was studied. The result suggeste that the relative standard deviation of absorbance of blackcurrant sample in the each wavelength was the minimum up to 4.963% at 30℃. For the same blackcurrant, the change of absorbance average was not significant with the increase of scan number. The relative standard up to 4.963% was accorded with the requirement when the scan number was 6. The effect of variety and size of blackcurrant was not significant on absorbance average atα=0.05, while the influence of surface color on absorbance average was significant atα=0.05. Therefore, the near infrared detection model of nutritional component content can be constructed by the variety and size of blackcurrant.2. The optimum mathematic model of quantitative analysis for total acid, VC, total sugar and anthocyanin content was developed according to stepwise regression algorithm within multiple linear regression by SPSS17.0 software. The correlation coefficient of calibration for total acid was improved from 0.7093 optimized to 0.9280, and standard deviation of calibration was decreased from 0.7655% to 0.3727%, and correlation coefficient of external validation and RMSEP were 0.9660 and 0.25% respectively. The correlation coefficient of calibration for VC was improved from 0.8159 optimized to 0.9730, and standard deviation of calibration was decreased from 23.2mg/100g to 8.3mg/100g, and correlation coefficient of external validation and RMSEP were 0.9770 and 9.2mg/100g, respectively. The correlation coefficient of calibration for total sugar was improved from 0.6547 optimized to 0.9480, and standard deviation of calibration was decreased from 1.5773% to 0.55%, and correlation coefficient of external validation and RMSEP were 0.9723 and 0.46% respectively. The correlation coefficient of calibration for anthocyanin was improved from 0.8411 optimized to 0.9070, and standard deviation of calibration was decreased from 66.8mg/100g to 50.9mg/100g, and correlation coefficient of external validation and RMSEP were 0.9649 and 25.0mg/100g respectively.3. According to the mathematic model obtained, the forecast of nutritional component for six varieties of blackcurrant during the 13 harvesting periods from June 2010 to August 2010 was arrived by the adjustment of Slope/Bias Correction Method. The variables of nutritional component for different varieties during the harvesting period was obtained, and the optimal harvesting period for each variety was determined, which of Heifeng between July 22 and July 25, Bajila between August 5 and August 11, Yard between July 17 and July 20, 94-4-13 about July 21, Daisha between August 1 and August 4, and Browder between July 8 and July 14. According to the internal nutrition quality for six varieties of blackcurrant, the sequence was 94-4-13> Browder>Daisha>Bajila>Heifeng>Yard.4. To solve the problem of blackcurrant adhesion in the image, a new image segmentation algorithm was presented in this research and can succeed in dividing the adhesive blackcurrant. On this basis, according to the requirement, the characteristic parameters of the area, perimeter, major axis length, minor axis length, elongation, rectangular degrees, circularity, and the average of H, S, V, R, G and B, as well as the value G/B, G/R and R/B for Heifeng, Bajila, Yard, 94-4-13, Daisha and Browder were defined and extracted, and the relation between 16 characteristic parameters and the nutritional quality, including total acid, VC, total sugar and anthocyanin content, were studied. The result suggeste that there is a certain correlation between nutritional component and appearance quality, but the correlation was poor.5. The Variety of blackcurrant can be identified based on BP neural network, and it was known by the validation experiment that the correct recognition rates for Heifeng, Bajila, Yard, 94-4-13, Daisha and Browder were 92%, 100%, 94%, 98%, 94% and 96%, respectively. The correct recognition rates were all accurate, and the recognition method can be used to detect the variety of blackcurrant in the actual produce. In the evaluation and classification of quality for different varieties of blackcurrant, the classification standard was provided. It was proven according to validation experiment that there were significant differences among the comprehensive scores of the different classifications for each variety of blackcurrant atα=0.05. Therefore, the classification method is available.In this research, the quantitative analysis model of nutrition, variety identification and the quality classification method of comprehensive evaluation obtained can be used to provide theoretical basis and guidance for quality management during the production process of berry.