• Spectroscopy and Spectral Analysis
  • Vol. 42, Issue 9, 2836 (2022)
Raw spectra of plum fruit
Fig. 1. Raw spectra of plum fruit
The probability distribution of SSC (a) and TA (b) content for plum samples
Fig. 2. The probability distribution of SSC (a) and TA (b) content for plum samples
Quantitative results of PLS model for (a) SSC and (b) TA in plum samples
Fig. 3. Quantitative results of PLS model for (a) SSC and (b) TA in plum samples
Quantitative results of BP-ANN model for (a) SSC and (b) TA in plum samples
Fig. 4. Quantitative results of BP-ANN model for (a) SSC and (b) TA in plum samples
指标名称波段/cm-1校正集交叉验证因子数剔除样品
rRMSECrRMSECV
SSC/%4 000~9 0000.890 1
0.891 1
1.01
1.01
0.876 3
0.877 5
1.06
1.06
8
8
剔除7个
剔除8个
4 000~8 8520.901 6
0.901 3
0.904 8
0.96
0.96
0.94
0.888 7
0.887 3
0.893 2
1.01
1.02
0.99
8
8
8
剔除7个
剔除8个
剔除10个
4 000~8 0000.886 21.030.871 81.088剔除6个
4 000~6 0000.883 9
0.884 5
1.04
1.03
0.872 9
0.873 0
1.07
1.07
7
7
剔除7个
剔除8个
6 000~9 0000.888 51.020.872 81.087剔除6个
TA/%4 000~9 0000.821 20.890.811 40.906剔除5个
4 000~6 5230.827 50.870.818 70.885剔除5个
4 605~9 0000.822 30.880.812 20.906剔除5个
4 605~6 5230.833 00.860.824 50.875剔除5个
5 605~9 0000.824 80.880.812 30.906剔除5个
5 605~6 5230.821 70.880.799 60.925剔除6个
6 410~7 6000.818 50.890.810 20.905剔除6个
Table 1. Comparison on performance of different model bands
指标名称样品集样品数数值范围平均值标准偏差
SSC
/%
校正集
预测集
382
191
5.20~16.00
5.40~14.90
10.16
10.19
2.23
2.14
TA/%校正集
预测集
267
130
6.15~13.45
6.44~12.68
9.88
9.84
1.55
1.52
Table 2. Statistics of SSC, TA in plum for calibration set and prediction set
光谱预处理校正集预测集因子数
RcRMSECRpRMSEP
原始光谱0.903 90.960.874 01.048
消除常数偏移量0.913 50.920.875 91.029
减去一条直线0.909 00.940.874 81.038
SNV0.911 40.930.874 01.048
最大-最小归一化0.903 40.970.863 31.088
MSC校正0.914 40.910.878 51.008
一阶微分+5点平滑0.876 91.080.876 61.014
一阶微分+17点平滑0.844 01.200.833 01.184
一阶微分+25点平滑0.862 21.140.874 71.035
二阶微分+17点平滑0.869 81.110.795 51.294
二阶微分+25点平滑0.842 61.210.812 21.244
一阶微分+减去一条
直线+5点平滑
0.904 80.950.874 11.044
一阶微分+SNV+25点平滑0.905 20.950.876 01.028
一阶微分+MSC+25点平滑0.905 60.950.875 41.027
Table 3. Performance comparison with the SSC model of different spectral pretreatment methods
光谱预处理校正集预测集因子数
RcRMSECRpRMSEP
原始光谱0.840 30.850.807 20.895
消除常数偏移量0.795 20.950.794 80.924
减去一条直线0.830 60.870.796 30.925
SNV0.843 50.850.819 30.874
最大-最小归一化0.826 30.880.807 70.895
MSC校正0.841 70.840.814 00.884
一阶微分+17点平滑0.841 00.840.813 90.884
一阶微分+25点平滑0.815 50.910.812 00.884
二阶微分+17点平滑0.841 20.840.760 80.984
二阶微分+25点平滑0.825 30.880.790 60.934
一阶微分+减去一条
直线+5点平滑
0.815 10.910.774 20.964
一阶微分+SNV+9点平滑0.860 30.800.819 60.866
一阶微分+SNV+25点平滑0.840 70.850.816 80.875
一阶微分+MSC+25点平滑0.809 20.920.775 30.954
Table 4. Performance comparison with the TA model of different spectral pretreatment methods
参数SSCTA
输入层函数tansigtansig
隐含层函数logsiglogsig
输出层函数purelintansig
训练函数trainlmtraingdx
显示间隔5050
学习速率0.050.05
训练目标1×10-31×10-3
Table 5. Parameters of the BP-ANN models for SSC, TA in plums
隐含层节点数校正集预测集
RcRMSECRpRMSEP
100.935 10.840.719 31.15
110.945 10.820.671 21.25
120.925 40.850.660 21.29
130.950 80.810.755 31.15
140.955 70.810.782 01.04
150.960 40.800.797 91.04
160.964 10.790.854 01.02
170.964 30.790.871 91.01
180.976 70.750.889 70.99
190.967 20.780.713 21.16
200.965 90.780.675 91.22
Table 6. Influence of nodes in hidden layer on quantitative results of SSC in plum
隐含层节点数校正集预测集
RcRMSECRpRMSEP
100.941 00.660.806 40.88
110.947 90.650.787 70.91
120.959 10.640.817 70.86
130.958 40.640.800 60.91
140.962 10.630.806 00.87
150.956 70.640.830 70.85
160.955 20.640.795 40.91
170.956 70.640.806 20.87
180.960 80.630.877 30.84
190.974 30.620.897 70.83
200.968 40.630.867 60.84
Table 7. Influence of nodes in hidden layer on quantitative results of TA in plum
检测指标建模方法校正集预测集
RcRMSECRpRMSEP
SSC/%PLS
BP-ANN
0.914 4
0.976 7
0.91
0.75
0.878 5
0.889 7
1.00
0.99
TA/%PLS
BP-ANN
0.860 3
0.974 3
0.80
0.62
0.819 6
0.897 7
0.86
0.83
Table 8. Quantitative results of the established models using PLS and BP-ANN methods for SSC, TA in plum