Spectroscopy and Spectral Analysis, Volume. 42, Issue 8, 2380(2022)
Research on Protein Powder Adulteration Detection Based on Hyperspectral Technology
Figures & Tables(10)
Fig. 2. Spectral curves of protein powders adulterated with different substances
(a): Original spectra of all samples; (b): Spectra of protein powders adulterated by 20% other powders
Fig. 3. The average spectra of three types of adulterated samples with different blend rations
(a), (b), (c) represent the average spectral curves of protein powder samples mixed with corn flour, rice flour and wheat flour, respectively
Fig. 4. Scatter plots of prediction models of three types of adulterated samples by using CARS-LSSVM
(a), (b), (c) are the scatter plots of prediction models for the protein powder mixed with corn flour, rice flour and wheat flour, represent
Table 1. Mass percentages and the number of three types of adulterated samples
View tableView in ArticleTable 1. Mass percentages and the number of three types of adulterated samples
质量分数 样本个数 质量分数 样本个数 5% 8 35% 8 10% 8 40% 8 15% 8 45% 8 20% 8 50% 8 25% 8 55% 8 30% 8 60% 8
Table 2. Specific parameter settings of hyperspectral system
View tableView in ArticleTable 2. Specific parameter settings of hyperspectral system
名称 参数 相机曝光时间 6 ms 光谱分辨率 10 nm 光谱范围 397.5~1 014 nm 位移平台移动速度 前进1 m·s-1,回退2.5 m·s-1
Table 3. Comparison of accuracy rates of the three models based on the original and preprocessed full-band spectra
View tableView in ArticleTable 3. Comparison of accuracy rates of the three models based on the original and preprocessed full-band spectra
预处理 建模方法 掺假物/% 整体/% RP RMSEP 无 玉米粉 大米粉 小麦粉 Raw PCR 100 87.5 75.0 58.3 80.2 0.949 3 0.351 5 SG 100 91.7 87.5 75.0 88.5 0.957 6 0.322 2 Normalize 100 95.8 100 91.7 96.9 0.979 9 0.223 0 MSC 100 95.8 100 95.8 97.9 0.983 6 0.201 9 Baseline 100 79.2 79.2 45.8 76.1 0.890 5 0.508 7 SNV 95.8 95.8 79.2 70.8 85.4 0.962 7 0.302 6 Raw BPNN 58.3 33.3 79.2 16.7 46.9 0.733 4 0.772 5 SG 91.7 37.5 45.9 41.7 54.2 0.699 2 0.880 6 Normalize 100 36.4 79.2 37.5 63.3 0.842 1 0.606 4 MSC 91.7 75.0 62.5 45.8 68.8 0.844 0 0.603 6 Baseline 16.7 87.5 54.2 45.8 51.1 0.848 4 0.594 6 SNV 91.7 83.3 66.7 37.5 69.8 0.872 5 0.548 1 Raw RF 100 91.7 100 100 97.9 0.996 3 0.023 9 SG 100 95.8 100 100 99.0 0.996 4 0.023 7 Normalize 100 100 100 91.7 97.9 0.993 3 0.032 7 MSC 100 100 100 100 100 0.997 9 0.018 9 Baseline 100 95.8 100 83.3 94.8 0.988 4 0.044 7 SNV 100 100 95.8 100 99.0 0.996 5 0.023 9
Table 4. The prediction results of the LSSVM (RBF_kernel) models based on the original and preprocessed full-band spectra
View tableView in ArticleTable 4. The prediction results of the LSSVM (RBF_kernel) models based on the original and preprocessed full-band spectra
掺假物 预处理 参数 RP RMSEP γ σ2 玉米粉 无 4 464 077.790 8 921.039 0 0.996 1 0.017 8 SG 2 026.247 6 591.953 4 0.992 8 0.023 2 Normalize 590.163 6 2 216.417 6 0.983 6 0.036 5 MSC 615.210 9 4 394.301 7 0.989 6 0.027 8 Baseline 501.078 8 913.596 5 0.971 9 0.050 0 SNV 643.856 9 4 555.332 4 0.989 6 0.027 9 大米粉 无 20 531.236 9 2 779.792 9 0.993 3 0.024 3 SG 48 841.081 5 3 430.185 6 0.989 7 0.030 4 Normalize 82 075.203 5 9 667.740 7 0.990 8 0.030 7 MSC 3 603.650 4 4 913.051 1 0.983 9 0.040 5 Baseline 54 097.003 8 2 042.103 4 0.996 1 0.018 4 SNV 2 629.220 2 4 911.529 0 0.984 4 0.339 9 小麦粉 无 838 972.423 9 355.791 2 0.997 3 0.015 0 SG 290 053.071 8 320.415 8 0.996 9 0.016 1 Normalize 149 529.038 6 802.730 2 0.997 8 0.013 8 MSC 702.265 6 199.747 0 0.992 0 0.027 5 Baseline 1 031 440.655 2 12 291.456 5 0.990 9 0.027 1 SNV 708.193 4 243.926 6 0.990 4 0.029 9
Table 5. Screening results of two band screening methods
View tableView in ArticleTable 5. Screening results of two band screening methods
掺假物 方法 个数 波长/nm 玉米粉 SPA 12 400.8, 404.2, 407.6, 414.3, 424.4, 434.6, 492.3, 536.9, 704.0, 969.9, 988.3, 1 010.3 CARS 26 400.8, 465.1, 482.1, 485.5, 526.6, 592.1, 609.5, 626.9, 647.8, 682.9, 686.4, 693.4, 704.0, 707.5, 746.4, 750.0, 753.5, 774.9, 814.2, 821.3, 828.5, 875.3, 922.5, 959.0, 988.3, 1 006.7 大米粉 SPA 10 414.3, 427.8, 835.7, 977.3, 988.3, 995.6, 999.3, 1 003.0, 1 010.3, 1 014.0 CARS 41 397.5, 444.7, 451.5, 475.3, 482.1, 488.9, 495.8, 512.8, 530.0, 536.9, 547.2, 588.6, 595.6, 661.8, 665.3, 668.8, 672.3, 675.8, 679.3, 686.4, 711.0, 732.2, 764.2, 799.9, 803.4, 824.9, 839.3, 857.3, 875.3, 879.0, 886.2, 893.4, 918.8, 929.8, 933.4, 937.1, 948.0, 962.6, 980.9, 991.9, 995.6 小麦粉 SPA 13 400.8, 404.2, 407.6, 417.7, 421.1, 444.7, 468.5, 630.3, 969.9, 991.9, 999.3, 1 006.7, 1 010.3 CARS 47 407.6, 421.1, 434.6, 468.5, 471.9, 482.1, 488.9, 492.3, 512.8, 526.6, 533.4, 540.3, 547.2, 585.2, 623.4, 630.3, 647.8, 679.3, 686.4, 732.2, 735.8, 757.1, 760.6, 764.2, 771.3, 778.4, 782.0, 785.6, 799.9, 810.6, 824.9, 828.5, 832.1, 835.7, 893.4, 904.3, 911.6, 918.8, 929.8, 933.4, 955.3, 969.9, 980.9, 988.3, 995.6, 999.3, 1 003.0
Table 6. Prediction results of adulterant content in protein powder using LSSVM (RBF_kernel) models based on different characteristic wavelength screening methods
View tableView in ArticleTable 6. Prediction results of adulterant content in protein powder using LSSVM (RBF_kernel) models based on different characteristic wavelength screening methods
模型类型 掺假物 参数 RP RMSEP γ σ2 玉米粉 1 614.769 5 191.968 1 0.989 0 0.032 8 SPA-LSSVM 大米粉 2 093.339 1 29.057 6 0.986 0 0.039 3 小麦粉 55 737.505 1 911.686 1 0.997 9 0.014 1 玉米粉 276 020.435 6 1 121.642 2 0.991 0 0.025 3 CARS-LSSVM 大米粉 5 590.697 5 175.167 7 0.994 6 0.022 7 小麦粉 22 773.349 2 288.292 7 0.999 1 0.008 8
Tools
Get Citation
Copy Citation Text
Bin LI, Hai YIN, Feng ZHANG, Hui-zhen CUI, Ai-guo OUYANG. Research on Protein Powder Adulteration Detection Based on Hyperspectral Technology[J]. Spectroscopy and Spectral Analysis, 2022, 42(8): 2380
Paper Information
Category: Orginal Article
Received: Jul. 7, 2021
Accepted: --
Published Online: Mar. 17, 2025
The Author Email: LI Bin (libingioe@126.com)
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20