1 材料与方法
1.1 苜蓿干草样本的采集
表1 苜蓿干草样本信息Table 1 Information of alfalfa hay samples |
| 地点 Locations | 茬次 Cutting number | 生长期 Growth period | 样本数量 Number of samples |
|---|---|---|---|
| 酒泉 Jiuquan | 一茬 | 初花期 | 83 |
| 一茬 | 盛花期 | 15 | |
| 二茬 | 初花期 | 28 | |
| 二茬 | 盛花期 | 9 | |
| 三茬 | 初花期 | 14 | |
| 张掖 Zhangye | 一茬 | 初花期 | 60 |
| 二茬 | 初花期 | 36 | |
| 兰州 Lanzhou | 一茬 | 初花期 | 53 |
| 二茬 | 初花期 | 21 | |
| 三茬 | 现蕾期 | 6 | |
| 三茬 | 初花期 | 7 | |
| 庆阳 Qingyang | 一茬 | 初花期 | 65 |
| 二茬 | 初花期 | 17 | |
| 二茬 | 盛花期 | 8 | |
| 三茬 | 现蕾期 | 6 | |
| 三茬 | 初花期 | 6 | |
| 天水 Tianshui | 一茬 | 初花期 | 48 |
| 二茬 | 初花期 | 11 |
1.2 近红外光谱数据采集
1.3 苜蓿干草成分测定
1.4 近红外光谱预测模型的构建与验证
2 结果与分析
2.1 苜蓿干草中NDF和ADF含量及RFV的分布
表2 苜蓿干草NDF和ADF含量及RFV样本集分类结果Table 2 Sample set classification results of contents of NDF and ADF as well as RFV in alfalfa hay |
| 项目 Items | 数据集 Data sets | 样本数量 Number of samples | 最大值 Maximum/ % | 最小值 Minimum/ % | 平均值 Mean/% | 中位值 Median/% | 标准偏差 Standard deviation/% |
|---|---|---|---|---|---|---|---|
| 中性洗涤纤维 NDF | 所有数据 All data | 493 | 89.91 | 26.97 | 44.79 | 44.42 | 5.71 |
| 校正集数据 Calibration set data | 370 | 89.91 | 26.97 | 44.91 | 44.49 | 5.85 | |
| 预测集数据 Prediction set data | 123 | 75.36 | 29.11 | 44.44 | 44.15 | 5.28 | |
| 酸性洗涤纤维 ADF | 所有数据 All data | 493 | 57.76 | 19.42 | 33.43 | 32.91 | 4.25 |
| 校正集数据 Calibration set data | 370 | 57.76 | 19.42 | 33.57 | 33.01 | 4.27 | |
| 预测集数据 Prediction set data | 123 | 50.17 | 20.85 | 33.03 | 32.53 | 3.98 | |
| 相对饲喂价值 RFV | 所有数据 All data | 493 | 254.43 | 45.42 | 133.08 | 132.59 | 21.80 |
| 校正集数据 Calibration set data | 370 | 254.43 | 45.42 | 132.56 | 132.30 | 21.38 | |
| 预测集数据 Prediction set data | 123 | 231.60 | 61.50 | 134.64 | 134.39 | 21.65 |
2.2 近红外光谱模型的构建
图2 不同预处理方法建立模型中SECV值随不同因子数的变化情况a:无预处理建模 modeling without pretreatment;b:SNV法处理后建模 modeling after SNV method treatment;c:Detrend法处理后建模 modeling after Detrend method treatment;d:SNV+Detrend法处理后建模 modeling after SNV+Detrend method treatment;e:SNV+Detrend+一阶导数法处理后建模 modeling after SNV+Detrend+first derivative method treatment。 Fig.2 Changes of SECV value with factor number in different models with different pretreatment methods |
表3 不同预处理方法苜蓿干草NDF和ADF含量及RFV校正结果Table 3 Calibration results of contents of NDF and ADF as well as RFV in alfalfa hay with different pretreatment methods |
| 项目 Items | 重复次数 Repeat times | 保留 Keep | 剔除 Reject | 因子数 Factors | 校正标 准偏差 SEC | 校正决 定系数 | 交叉验 证标准 偏差 SECV | 交叉验证 决定系数 | 参考值 标准差 SDref/% |
|---|---|---|---|---|---|---|---|---|---|
| 中性洗涤纤维NDF | |||||||||
| 无预处理 Non-pretreatment | 4 | 358 | 12 | 15 | 0.654 | 0.975 | 0.783 | 0.960 | 4.117 |
| SNV法SNV method | 2 | 363 | 7 | 15 | 0.511 | 0.986 | 0.591 | 0.978 | 4.367 |
| Detrend法 Detrend method | 4 | 358 | 12 | 15 | 0.585 | 0.980 | 0.689 | 0.969 | 4.096 |
| SNV+Detrend法 SNV+Detrend method | 4 | 360 | 10 | 15 | 0.454 | 0.988 | 0.584 | 0.978 | 4.111 |
| SNV+Detrend+一 阶导数法 SNV+Detrend+first derivative method | 2 | 364 | 6 | 13 | 0.528 | 0.985 | 0.700 | 0.972 | 4.271 |
| 酸性洗涤纤维ADF | |||||||||
| 无预处理 Non-pretreatment | 4 | 350 | 20 | 15 | 0.449 | 0.981 | 0.514 | 0.970 | 3.251 |
| SNV法SNV method | 3 | 363 | 7 | 15 | 0.365 | 0.990 | 0.433 | 0.983 | 3.600 |
| Detrend法 Detrend method | 4 | 353 | 17 | 15 | 0.424 | 0.984 | 0.511 | 0.974 | 3.354 |
| SNV+Detrend法 SNV+Detrend method | 3 | 363 | 7 | 15 | 0.350 | 0.991 | 0.444 | 0.983 | 3.597 |
| SNV+Detrend+ 一阶导数法 SNV+Detrend+first derivative method | 2 | 363 | 7 | 8 | 0.553 | 0.976 | 0.672 | 0.960 | 3.591 |
| 相对饲喂价值RFV | |||||||||
| 无预处理 Non-pretreatment | 4 | 349 | 21 | 15 | 2.269 | 0.975 | 3.028 | 0.966 | 16.540 |
| SNV法SNV method | 4 | 349 | 21 | 15 | 2.158 | 0.981 | 2.521 | 0.972 | 15.719 |
| Detrend法 Detrend method | 4 | 350 | 20 | 15 | 2.383 | 0.979 | 2.939 | 0.967 | 16.386 |
| SNV+Detrend法 SNV+Detrend method | 4 | 349 | 21 | 14 | 1.979 | 0.984 | 2.305 | 0.976 | 15.500 |
| SNV+Detrend+ 一阶导数法 SNV+Detrend+first derivative method | 4 | 352 | 18 | 10 | 2.528 | 0.976 | 3.051 | 0.963 | 16.170 |
图3 不同预处理方法校正集NDF和ADF含量及RFV参考值与模型预测值之间的相关关系a:无预处理建模 modeling without pretreatment;b:SNV法处理后建模 modeling after SNV method treatment;c:Detrend法处理后建模 modeling after Detrend method treatment;d:SNV+Detrend法处理后建模 modeling after SNV+Detrend method treatment;e:SNV+Detrend+一阶导数法处理后建模 modeling after SNV+Detrend+first derivative method treatment。 Count:计数;SEC:校正标准偏差 standard error of calibration; :校正决定系数 determination coefficient of calibration。 Fig.3 Correlations between reference values of contents of NDF and ADF as well as RFV in calibration set and predicted values in model under different pretreatment methods |
2.3 校正模型的验证
表4 不同预处理方法苜蓿干草NDF和ADF含量及RFV模型预测集预测结果Table 4 Predicted results of prediction set in model of contents of NDF and ADF as well as RFV in alfalfa hay with different pretreatment methods |
| 项目 Items | 化学实测 平均值 Chemical measured mean/% | 模型预测 平均值 Model prediction mean/% | 差值 Difference/ % | 斜率 Slope | 预测标 准偏差 SEP/% | 预测决 定系数 | 相对分 析误差 RPD |
|---|---|---|---|---|---|---|---|
| 中性洗涤纤维NDF | |||||||
| 无预处理Non-pretreatment | 44.441 | 44.380 | 0.061 | 1.004 | 1.147 | 0.970 | 3.389 |
| SNV法SNV method | 44.256 | 0.185 | 1.161 | 1.755 | 0.950 | 2.488 | |
| Detrend法Detrend method | 44.752 | 0.311 | 0.746 | 2.814 | 0.932 | 1.456 | |
| SNV+Detrend法 SNV+Detrend method | 44.060 | 0.381 | 1.227 | 3.556 | 0.744 | 1.156 | |
| SNV+Detrend+一阶导数法 SNV+Detrend+first derivative method | 44.353 | 0.088 | 1.143 | 1.305 | 0.977 | 3.273 | |
| 酸性洗涤纤维ADF | |||||||
| 无预处理Non-pretreatment | 33.033 | 33.565 | 0.532 | 0.493 | 4.902 | 0.730 | 0.663 |
| SNV法SNV method | 33.120 | 0.087 | 0.950 | 0.663 | 0.984 | 5.430 | |
| Detrend法Detrend method | 34.655 | 0.622 | 0.473 | 5.212 | 0.720 | 0.642 | |
| SNV+Detrend法 SNV+Detrend method | 32.823 | 0.210 | 1.071 | 2.183 | 0.796 | 1.648 | |
| SNV+Detrend+一阶导数法 SNV+Detrend+first derivative method | 33.085 | 0.052 | 0.995 | 0.760 | 0.975 | 4.694 | |
| 相对饲喂价值RFV | |||||||
| 无预处理Non-pretreatment | 134.639 | 132.613 | 2.026 | 0.815 | 10.893 | 0.836 | 1.518 |
| SNV法SNV method | 133.777 | 0.862 | 1.033 | 5.679 | 0.944 | 2.770 | |
| Detrend法Detrend method | 131.850 | 2.789 | 0.667 | 15.720 | 0.755 | 1.042 | |
| SNV+Detrend法 SNV+Detrend method | 133.418 | 1.221 | 1.056 | 6.059 | 0.939 | 2.558 | |
| SNV+Detrend+一阶导数法 SNV+Detrend+first derivative method | 133.456 | 1.183 | 1.059 | 6.236 | 0.935 | 2.593 |
