Vis/NIR spectroscopy is considered one of the methods used to quickly and non-destructively determine the quality of agricultural products. This method can check the internal and external properties of agricultural products indirectly. On the other hand, a large amount of sugar is produced annually in the world using sugar beet, indicating the high importance of this product. The most important quality component of sugar beet is its sucrose content. Conventional methods for measuring the sucrose content in sugar beet are time-consuming and need consuming chemical substances such as lead acetate and aluminum sulfate, regarded as dangerous waste factors. In this study, the feasibility on Vis/NIR spectroscopy was assessed in the fast determination of sucrose content in sugar beet juice. A total of 362 sugar beet juice samples were collected from Isfahan Sugar Company. In the first step, the sucrose content of the samples was measured by the conventional method. Next, the samples were transferred to the non-destructive laboratory of the Biosystems Engineering Department for spectrometry. The Vis/NIR spectroscopy system used in this study included a light source (2 halogen lamps), optical fiber, dimmer, Vis/SWNIR and NIR spectrometers, and a portable computer. The Vis/SWNIR spectrometer was a CCD-type PDA spectrometer with a resolution of 0.2 nm, and the NIR spectrometer was a point-sensor type with a resolution of 9 nm. The spectra of the samples were collected in three spectral ranges: 400-1100, 950-1650, and 400-1650 nm. After collecting the spectral data, the data were analyzed and eva‎luated using Unscramble x software. To prepare the data, various pre-processing methods were applied and the best pre-processing was used for modeling in each spectral range. The partial least square (PLS) method was used to develop sucrose prediction models. The results showed that the first derivative preprocessing was the best pretreatment for the development of sucrose prediction models in all three spectral ranges. The best model in the spectral range of 400-1100 nm resulted in a correlation coefficient (rp) of 0.777, root means square error (RMSEP) of 0.87, and residual predictive deviation (RPD) of 2.18, indicating a very good ability of sucrose predictive model. The best model in the spectral range of 950-1650 nm led to an rp of 0.573, RMSEP of 1.10, and RPD of 1.72, indicating the acceptable ability of the built model. The best PLS model in sucrose prediction, however, was obtained by the combination of data from two spectrometers (spectral range 400-1650 nm) as the input to the models. The corresponding PLS model along with the first derivative preprocessing was able to predict the sucrose content with an rp of 0.817, RMSEP of 0.82, and RPD of 2.32 with a very good performance.

احمد ميره‌اي

جواد كرامت

مرتضي صادقي، محمد مهدي مجيدي