PREDICTION AND ANALYSIS OF STRAWBERRY SUGAR CONTENT BASED ON PARTIAL LEAST SQUARES PREDICTION MODEL
S. Liu1,2, H. Xu1,2*, J. Wen1, W. Zhong1 and J. Zhou1
1College of Engineering, Huazhong Agricultural University, Wuhan 430070; 2Key Laboratory of Agricultural Equipment in Mid-lower Yangtze River, Ministry of Agriculture and rural affairs
Corresponding author’s email: xhm790912@163.com
ABSTRACT
Non-destructive detection of fresh strawberry fruits is one of the research hotspots in the quality detection of agricultural products. Here, the hyperspectral data of Toyonoka and Jingyao strawberry were first collected by a hyperspectral imaging system. The reflectivity of the original spectra was corrected. Then, the Toyonoka strawberry spectra were preprocessed by the combination of moving-average data smoothing technique, 2 order derivative and multiplicative scatter correction; similarly, the Jingyao strawberry spectra were pretreated by the combination of Savizky-Golay smoothing technique, 2 order derivative and standard normal transformation method. Finally, correlation coefficient method and spectral difference analysis technique were combined to reduce the dimensions of the pretreated spectra and extract the characteristic wavelengths. Based on the above results, a partial least-squares prediction model of strawberry sugar was constructed. The prediction results with the model showed that in the calibration set, the correlation coefficient Rc was 0.8776 and 0.9004, and the standard deviation was 0.5100 and 0.7516 for Toyonoka and Jingyao strawberry, respectively; and in the validation set, the correlation coefficient Rp was 0.7708 and 0.8053, and the standard deviation was 0.7365 and 0.9947 for Toyonoka and Jingyao strawberry, respectively. The prediction effect and stability of the partial least-squares prediction model for Jingyao strawberry were superior to those for Toyonoka strawberry. Our results provide some references for the on-line and non-destructive detection of fruits and vegetables.
Key words: strawberries; sugar content prediction; hyperspectral technique; characteristic information extraction; partial least-squares method.
|