Hyperspectral Determination of Reducing Sugar in Potatoes Based on CARS

Wei Jiang, Junlong Fang, Shuwen Wang, Runtao Wang
2016 International Journal of Hybrid Information Technology  
It usually contains a large amount of redundant information to use the hyperspectral information to create a model, which will increase the difficulty of the model analysis. Therefore, it's so important to select the characteristic wavelength in an effective and quick way. This study is proposed by using the competitive adaptive reweighed sampling (CARS) to select the characteristic wavelength for detecting the reducing sugar content in the potatoes. In that experiment a total of 238 samples
more » ... l of 238 samples are prepared. Among them, 190 samples are selected as the calibration set and 48 samples as the validation set. The performance of CARS is compared with full spectrum and classical variable extraction methods such as Monte Carlo uninformative variable elimination (MC-UVE), genetic algorithm (GA) and moving window partial least squares (MWPLS). Experimental results show that the band screened by algorithm CARS has the best effect, compared to full spectrum modeling, the wavelength of the model reduces from 203 to 33, the model validation set coefficient R 2 increases from 0.8464 to 0.8965, and the root mean square error prediction (RMSEP) decreases from 0.0758 to 0.0416. The results demonstrate that it is feasible to detect the reducing sugar content of potatoes by using CARS combined with hyperspectral imaging. key variables before the quantitative analysis of the internal quality of agricultural products by using hyperspectral data. Studies have shown that the model that is more easily to be explained and has more stable performance can be obtained via spectrum optimization [6, 7] . In this paper, the main task is to use competitive adaptive reweighed algorithm (CARS) to select the characteristic wavelength after obtaining spectral information of potato based on hyperspectral imaging technology by using potato as the research object, and also respectively create partial least squares ( PLS)model by comparing with extraction methods of the full spectrum and other variables , such as Monte Carlo uninformative variable elimination (MCUVE), genetic algorithm( GA) and move window partial least squares( MWPLS) , and use the validation set to confirm the validity of the model through comprehensive comparison of various variable selection methods in the prediction results of reducing sugar content in the potato so as to obtain the optimal application of various variable selection methods of hyperspectral in quantitative analysis of the quality. increase of smooth points, the performance of PLS model is gradually decreased. After the maximum normalized and orthogonal signal correction, the performance of PLS model is slightly decreased, while it is the worst after the pretreatment of baseline correction, and the RMSEP is 0.0791.
doi:10.14257/ijhit.2016.9.9.04 fatcat:xpw5wbbr2zcn7oalcgzolppmpq