XU Sai, LU Huazhong, WANG Xu, QIU Guangjun, LIANG Xin, WANG Chen
To find the relationship between external vision features and internal quality detection of pomelo, and provide a rough detection method, twenty vision features (vertical and transverse diameters, ratio of vertical and transverse diameters, area, first order of R, G and B gray values, second-order of R, G and B gray values, third order of R, G and B gray values, H, V, and S gray values, correlation, contrast, energy, and roughness, which were represented by F1 to F20) and 3 major internal quality parameters (granulation degree, total soluble solid (TSS) and water content) were detected and analyzed. The results showed that the external feature could be well described by machine vision technology. The linear correlation between whole vision features and internal quality parameters was insignificant. However, the nonlinear correlation between vision features and internal quality parameters was significant. Thereinto, the nonlinear correlation between vision features and granulation degree were insignificant except F5, F8, F9, F10, F11, F12, F13, F14, F15, F17, F19 and F20. The nonlinear correlation between vision features (F5, F8, F9, F10, F11, F12, F13, F14, F15 and F17) and TSS/water content were significant, the nonlinear correlation between other vision features and TSS/water were extremely significant. The granulation degree of pomelo cannot be detected by whole vision features or only significant correlated vision features. However, TSS and water content of pomelo could be roughly well detected by vision features, and the detection effects of whole vision features based were better than only extremely significant features based. Thus, machine vision features not only provide supply information for internal quality detection by other methods, but also can be applied to construct a low-cost and roughly nondestructive detection method for pomelo TSS and water content. This study also provide a reference and new ideas for the improvement of internal quality nondestructive detection of other fruits.
