1. College of Mechanical and Electrical Engineering, Tarim University, Alar 843300, China
2. Modern Agricultural Engineering Key Laboratory at Universities of Education Department of Xinjiang Uygur Autonomous Region，Alar 843300, China
3. College of Horticulture and Forestry, Tarim University, Alar 843300, China
4. Key Laboratory of Biological Resources Conservation and Utilization of Tarim Basin，Xinjiang Production and Construction Corps，Alar 843300, China
Abstract：Southern Xinjiang is the region with the largest apricot planting area in the country, with a wide variety of apricots. In the apricot fruit market, the quality and price of different varieties of apricots ware vary greatly, and the phenomenon of shoddy and uneven quality has seriously restricted the development of the apricot industry in Xinjiang. To investigate the feasibility of rapid detection of apricot varieties using visible/near-infrared spectroscopy, a non-destructive identification method for apricot varieties is set up based on the qualitative discriminant analysis of six varieties of apricots in the southern Xinjiang region by visible/near-infrared spectroscopy of samples with chemometrics methods. The spectral data of six apricot varieties (“Huang apricot”, “Ganlan apricot”, “Xiaobai apricot”, “Xiaomi apricot”, “Kumaiti” and “Xiaodiaogan apricot”) were collected in the range of 350~1 000 nm (VIS/NIR) and 1 000~2 500 nm (NIR) by the spectrometer. After deleting the obvious noise at the head of the original spectrum, the retained spectrum is processed using Savitzky-Golay (SG) convolution smoothing and multiple scatter correction (MSC) to eliminate the interference information in the spectrum. The original spectra are reduceddimension using principal component analysis (PCA), competitive adaptive re-weighted sampling (CARS), random frog (RF), successive projection algorithm (SPA), and linear discriminant analysis (LDA), naive Bayesian (NB), K-nearest neighbor (KNN), support vector machine (SVM) were combined with modeling the whole spectrum and the reduced spectrum. The results showed that the model based on full-spectral data has a comparatively accurate result, and the classification accuracy of the SVM model was 95.7% in the VIS/NIR range and 97.8% in the NIR range for the LDA model, which could achieve the discriminative analysis of different species of apricots. After the reduced-dimension of spectral data by PCA, CARS-SPA, RF-SPA and SPA, the model still maintained high classification accuracy, and the PCA-LDA model had 97.8% classification accuracy in the VIS/NIR range, and the RF-SPA-LDA model had 95.7% classification accuracy in the NIR range. The results of different models show that the classification effect of models in the VIS/NIR range was better than that in the NIR range; among the four dimensionality reduction methods, the PCA method has the best dimensionality reduction effect; among the four classifiers. The accuracy of LDA and SVM models is higher than that of NB and KNN models, which is more suitable for the identification of apricot varieties. The results show that the rapid and nondestructive identification of apricot varieties can be achieved based on the VIS/NIR range spectrum combined with principal component analysis and linear discriminant analysis method, which provides aninnovative way for online sorting and identifying apricot fruits.
[1] Baccichet I, Chiozzotto R, Spinardi A, et al. Scientia Horticulturae, 2022, 294: 110780.
[2] Gómez-Martínez H, Bermejo A, Zuriaga E, et al. Scientia Horticulturae, 2021, 277: 109828.
[3] HU Qian, CHEN Ying-ying, TAN Xiao-mei, et al(胡 倩，陈莹莹，谭晓梅，等). World Forestry Research(世界林业研究), 2022, 35(4): 20
[4] Fratianni F, D'Acierno A, Albanese D, et al. Foods, 2021, 11(1): 100.
[5] Yang Z L, Cai L W, Han L J, et al. Journal of Near Infrared Spectroscopy, 2021, 29(6): 313.
[6] Zeng J, Guo Y, Han Y Q, et al. Molecules, 2021, 26(3): 749.
[7] Sun J Y, Pang R C, Chen S S, et al. Journal of Innovative Optical Health Sciences, 2021, 14(6): 2130006.
[8] Zhao D Y, Xie D N, Yin F, et al. Remote Sensing, 2022, 14(10): 2420.
[9] Zhang H, Sun H F, Wang L, et al. Journal of Spectroscopy, 2018, 2018: 7652592.
[10] Li C H, Li L L, Wu Y, et al. Journal of Spectroscopy, 2018, 2018: 6935197.
[11] Cortés V, Cubero S, Blasco J, et al. Food and Bioprocess Technology, 2019, 12(6): 1021.
[12] Tong P J, Junliang K L, Wei T T, et al. Journal of Cereal Science, 2021, 102: 103322.
[13] HE Yong, ZHENG Qi-shuai, ZHANG Chu, et al(何 勇，郑启帅，张 初，等). Spectroscopy and Spectral Analysis(光谱学与光谱分析), 2019, 39(9): 2817.
[14] Dan S J. International Journal on Semantic Web and Information Systems, 2022, 18(1): 1.
[15] Qian L L, Li D W, Song X J, et al. Journal of Food Composition and Analysis, 2022, 105: 104203.
[16] CHENG Jie-hong, CHEN Zheng-guang(程介虹，陈争光). Chinese Journal of Analytical Chemistry(分析化学), 2021, 49(8): 1402.
[17] SA Ji-ming, JIANG He, XIE Kai-wen, et al(撒继铭，江 河，谢凯文，等). Acta Optica Sinica(光学学报), 2021, 41(15): 235.
