Study on the Detection and Discrimination of Damaged Jujube Based on Hyperspectral Data
YUAN Rui-rui1, WANG Bing2, LIU Gui-shan1*, HE Jian-guo1, WAN Guo-ling1, FAN Nai-yun1, LI Yue1, SUN You-rui1
1. School of Food & Wine, Ningxia University, Yinchuan 750021, China
2. School of Physics and Electronic-Electrical Engineering, Ningxia University, Yinchuan 750021, China
Abstract:Lingwu long jujube as Ningxia dominant characteristic jujube fruit. It has important economic and social value and scientific research significance. This paper has been lingwu long jujube as the research object. First, 60 intact jujubes images were collected Visible/near-infrared (Vis/NIR) using the hyperspectral imaging system. Damage tests were performed on 60 intact jujubes using the damaged device, and 60 damaged (internal bruising) jujube were obtained. The hyperspectral imaging system was used to collect the five time periods after damage (2, 4, 8, 12 and 24 h after damage) jujube spectral image. Region of interest (ROI) was extracted with ENVI software for the collected hyperspectral images of long jujube, and the average spectral value of intact long jujube and each time period long jujube were calculated. Then, the raw spectral data used Savitzky-golay smooth first derivatives (SG-1) and second derivatives (SG-2), standard normal variate (SNV) and de-trending, and the combined algorithms of SNV-SG-1, SNV-SG-2, de-trending-SG-1 and de-trending-SG-2 were pre-processed. The partial least squares-discriminant analysis (PLS-DA) classification model was established for the original spectrum and the pretreated spectrum. Finally, the optimal pre-processing spectral data were selected, and successive projection algorithm (SPA), interval random frog (IRF), uninformative variable elimination (UVE), variable combination population analysis (VCPA), interval variable iterative space shrinkage approach (IVISSA), IRF-SPA, UVE-SPA and IVISSA-SPA were used to select characteristic variables. The PLS-DA, linear discriminant analysis (LDA) and support vector machine (SVM) classification discriminant models were established for the selected feature variables. The results show that in the PLS-DA model based on the original spectral data, the accuracy of model calibration set and prediction set was 82.96% and 90%, respectively. After spectrum pretreatment, the SNV-SG-2-PLS-DA was obtained as the optimal classification discriminant model, and the accuracy of model calibration set and prediction set was 91.11% and 96.67%, respectively. In the classification model established by feature variables, the accuracy of the SNV-SG-2-UVE-PLS-DA model calibration set and prediction set were 86.3% and 94.44%, respectively. The accuracy of the SNV-SG-2-SPA-LDA model calibration set and prediction set were 86.3% and 83.33%, respectively. The accuracy of the SNV-SG-2-UVE-SVM model calibration set and prediction set were 77.78 and 71.11%, respectively. For the classification model, the classification results of the linear classification model (PLS-DA, LDA) were superior to those of the nonlinear classification model (SVM). The results of the linear classification model, PLS-DA was superior to LDA classification results, and PLS-DA could provide a better classification effect. The results show that the hyperspectral combined with the partial least squares-discriminant analysis model could effectively realize the rapid detection of the damage of lingwu long jujube of the change of time, providing a theoretical basis for the online detection of lingwu long jujube.
Key words:Lingwu long jujube; Hyperspectral; Partial least squares-discriminant analysis; Linear discriminant analysis; Support vector machine
袁瑞瑞,王 兵,刘贵珊, 何建国,万国玲,樊奈昀,李 月,孙有瑞. 高光谱数据对损伤长枣的检测判别[J]. 光谱学与光谱分析, 2021, 41(09): 2879-2885.
YUAN Rui-rui, WANG Bing, LIU Gui-shan, HE Jian-guo, WAN Guo-ling, FAN Nai-yun, LI Yue, SUN You-rui. Study on the Detection and Discrimination of Damaged Jujube Based on Hyperspectral Data. SPECTROSCOPY AND SPECTRAL ANALYSIS, 2021, 41(09): 2879-2885.
[1] Jiang W Q, Chen L H, Han Y R, et al. Scientia Horticulturae, 2020, 274:109667.
[2] Song L H, Cao B. Acta Horticulturae, 2016, 1116: 89.
[3] Wang Y T, Dai Y P, Xue J R, et al. EURASIP Journal on Image and Video Processing, 2017,2017: 34.
[4] Keresztes J C, Goodarzi M, Saeys W. Food Control, 2016, 66:215.
[5] Lee W, Kim M S, Lee H, et al. Journal of Food Engineering, 2014, 130: 1.
[6] CHI Qian,WANG Zhuan-wei,YANG Ting-ting, et al(迟 茜,王转卫,杨婷婷,等). Transactions of the Chinese Society for Agricultural Machinery(农业机械学报), 2015, 46(3): 235.
[7] Siedliska A, Baranowski P, Zubik M, et al. Postharvest Biology and Technology, 2018, 139: 115.
[8] Hu M H, Dong Q L, Liu B L. Computers and Electronics in Agriculture,2016, 122:19.
[9] Ye D, Sun L, Tan W, et al. Chemometrics and Intelligent Laboratory Systems, 2018, 177: 129.
[10] Zhang M, Li G H. International Journal of Food Properties, 2018, 21(1):1598.
[11] Fan S X, Li C Y, Huang W Q, et al. Postharvest Biology and Technology, 2017, 134:55.
[12] CHENG Li-juan, LIU Gui-shan, HE Jian-guo, et al(程丽娟,刘贵珊,何建国,等). Food Science(食品科学), 2019, 40(10):285.
[13] GUO Wen-chuan, DONG Jin-lei(郭文川,董金磊). Optics and Precision Engineering(光学精密工程), 2015, 23(6):1530.
[14] Guo Z M, Wang M M, Agyekum A A, et al. Journal of Food Engineering, 2020, 279:109955.
[15] Siedliska A, Baranowski P, Zubik M, et al. Postharvest Biology & Technology, 2018, 139: 115.