Spectroscopy and Spectral Analysis, Volume. 42, Issue 1, 282(2022)
A Model Construction Method of Spectral Nondestructive Detection for Apple Quality Based on Unsupervised Active Learning
Figures & Tables(7)
Fig. 1. Flow chart of spectral detecting method based on HAC-LLR training samples selecting strategy
Fig. 2. The average spectra of three cultivars apple samples harvestee from two years
Fig. 3. PLSR mdoel prediction results of SSC (a) and firmness (b) based on different sample selection algorithms under different datsets
Fig. 4. Normalized AUCs of the RMSE (a), the
Table 1. Statistics of quality reference for apple samples
View tableView in ArticleTable 1. Statistics of quality reference for apple samples
收获年份 品种 数量 SSC/% 硬度/N 均值 方差 最小值 最大值 均值 方差 最小值 最大值 GD 1 070 14.814 1.484 9.700 18.800 55.549 14.800 30.431 98.893 2009 JG 874 12.719 1.059 9.400 16.400 62.951 22.173 29.319 110.187 RD 1 078 11.558 1.160 8.100 15.400 58.911 15.099 29.497 89.980 GD 1131 13.232 1.270 9.600 17.500 63.280 16.698 29.434 100.994 2010 JG 1 087 12.545 1.532 8.700 17.300 63.405 19.015 27.700 105.340 RD 1 035 12.252 1.454 8.800 15.500 69.068 16.654 29.978 107.428
Table 2. The prediction results of PLSR models based on 200 samples from 2009 selected by four algorithms respectively
View tableView in ArticleTable 2. The prediction results of PLSR models based on 200 samples from 2009 selected by four algorithms respectively
品种 年份 算法 预测集 SSC 硬度 RMSE Rp RPD Dif./% RMSE Rp RPD Dif./% GD 2009 RS 0.824 0.873 2.00 5.9 9.057 0.825 1.765 3.3 KS 0.811 0.872 2.037 4.4 8.883 0.828 1.812 1.4 SPXY 0.842 0.873 1.971 8.0 9.399 0.822 1.719 6.8 HAC-LLR 0.775 0.882 2.122 8.760 0.835 1.831 JG 2009 RS 0.683 0.799 1.649 2.0 8.601 0.924 2.588 5.1 KS 0.726 0.764 1.556 7.9 8.353 0.928 2.672 2.3 SPXY 0.731 0.785 1.550 8.5 8.746 0.921 2.552 6.7 HAC-LLR 0.669 0.808 1.692 8.164 0.931 2.732 RD 2009 RS 0.802 0.776 1.505 8.5 9.257 0.814 1.702 4.3 KS 0.768 0.776 1.572 4.4 9.043 0.818 1.745 2.1 SPXY 0.795 0.781 1.513 7.7 9.097 0.817 1.732 2.6 HAC-LLR 0.734 0.800 1.650 8.856 0.827 1.778
Table 3. The prediction results of PLSR models based on 200 samples from 2010 selected by four algorithms respectively
View tableView in ArticleTable 3. The prediction results of PLSR models based on 200 samples from 2010 selected by four algorithms respectively
品种 年份 算法 预测集 SSC 硬度 RMSE Rp RPD Dif./% RMSE Rp RPD Dif./% GD 2010 RS 0.649 0.881 2.110 8.6 8.214 0.874 2.050 2.6 KS 0.632 0.887 2.177 6.2 8.096 0.882 2.076 1.2 SPXY 0.642 0.886 2.133 7.6 8.349 0.874 2.013 4.2 HAC-LLR 0.593 0.900 2.315 8.000 0.882 2.101 JG 2010 RS 0.759 0.887 2.165 5.0 11.083 0.828 1.764 4.2 KS 0.758 0.890 2.166 4.9 11.433 0.817 1.707 7.2 SPXY 0.742 0.901 2.211 2.8 11.194 0.829 1.743 5.2 HAC-LLR 0.721 0.899 2.279 10.615 0.837 1.845 RD 2010 RS 0.748 0.872 2.011 8.0 9.524 0.849 1.850 7.2 KS 0.714 0.880 2.104 3.6 9.229 0.850 1.917 4.2 SPXY 0.793 0.858 1.902 13.2 10.486 0.824 1.691 15.7 HAC-LLR 0.688 0.890 2.184 8.837 0.865 1.991
Tools
Get Citation
Copy Citation Text
Xiao-kang ZHAO, Xin ZHAO, Qi-bing ZHU, Min HUANG. A Model Construction Method of Spectral Nondestructive Detection for Apple Quality Based on Unsupervised Active Learning[J]. Spectroscopy and Spectral Analysis, 2022, 42(1): 282
Paper Information
Category: Research Articles
Received: Dec. 18, 2020
Accepted: --
Published Online: Mar. 31, 2022
The Author Email: Xiao-kang ZHAO (zhaoxk0211@163.com)
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20