| Potatoes contain 18 kinds of amino acids required in the human body and many kinds of trace elements, with a comprehensive and balanced, rich in nutrition. Different potatoes internal composition, similar to internal component content of potato classification can effectively improve the economic benefits of the potato. But the traditional test methods have the low efficiency, complex process, shortcomings and so on, not a detailed classification of potatoes in order to increase the economic benefits of the potato.Therefore to seek a rapid and nondestructive detection method is of great significance.Based on potato as the research object, this paper using near infrared hyperspectral imaging technology (900-1700nm) combined with chemometrics methods, the potato dry matter content, starch content, moisture content and protein content carries on the preliminary research, expected to seek a method of rapid detection of potato internal components and provide theoretical basis for potato internal composition online non-destructive testing.The main research content results as follows:(1) Dry matter content of potatoes was determined using near-infrared hyperspectral imaging technique. The original spectra were pretreated by multiplicative scatter correction, 8 optimal wavelengths were selected by regression coefficients of partial least-squares models in the spectral region between 990nm and 1630nm. Prediction models were built using Particle Swarm Optimization Algorithm optimizing Support Vector Machine method (PSO-SVM) and Partial Least Squares Regression method (PLSR) based on the optimal wavelengths. The results showed that prediction models based on PSO-SVM method in the optimal wavelengths are better than PLSR method for predicting the dry matter content in potatoes, its correlation coefficient and root mean square error of calibration and validation models are 0.94437,0.91977 and 0.15501,0.15690, respectively. Therefore, It’s feasible to determinate the dry matter content in potatoes using hyperspectral imaging technique.(2) With potatoes as the study object, summarizes a near-infrared (NIR) hyperspectral imaging technique in the range of 900-1700nm was investigated for non-destructive detection of starch content of potatoes. The hyperspectral images of potatoes over the spectral region between 900nm and 1700nm were acquired and obtained multiplicative scatter correction (MSC) wavelength. The important wavelengths were selected using regression coefficient (RC) and successive projections algorithm(SPA), and the models of multiple liner regression (MLR) and partial least-squares regression (PLSR) were then established using these feature wavelengths. The results suggested that the models of the important wavelength by SPA were better than regression coefficient method. And SPA-MLR model was superior to SPA-PLSR model to predict starch content of potatoes. The calibration coefficient of determination (Re) and root mean square error of calibration (RMSEC) were 0.9720391,0.3289996, respectively. The validation coefficient of determination (Rp) and root mean square error of prediction (RMSEP) were 0.982011,0.2488569, respectively.(3) A near-infrared (NIR) hyperspectral imaging technique in the range of 900-1700nm was investigated for non-destructive detection of moisture and protein content of potatoes. To moisture content, through the SPA method 22 feature band extracted, the paper was established characteristic band of SPA-PLSR, SPA-MLR, SPA-BP, and compared with the whole band PLSR model analysis.To protein content, through a competitive adaptive weight weighted algorithm (competitive the adaptive reweighted from, CARS), 18 feature band were extracted and built the CARS-PLSR, CARS-MLR, CARS-BP models under the characteristic band and compared with the whole band PLSR model analysis.The results show that the SPA - MLR model in optimal in water detection, the Rc, RMSEC, Rp, RMSEP of 0.842,0.491,0.831,0.589.CARS-BP model in optimal in protein detection, the Re, RMSEC, Rp.RMSEP of 0.851,0.126,0.821,0.115. |
PDF Full Text Request