贮备饲料近红外光谱模型快速预测青绿饲料营养成分含量

为快速检测饲料的营养成分,该研究利用贮备饲料的近红处技术(near-infrared,NIR)快速分析模型预测青绿饲料的营养成分含量。基于贮备饲料的NIR定标模型,将建模优化模式转移应用到青绿饲料的营养成分定量检测,以判断模型转移能力。在实验室环境下扫描并记录新鲜的青绿饲料样本和储存的贮备饲料样本的近红外反射光谱,利用230个贮备饲料样本进行光谱定标训练,以修正偏最小二乘(modified-partial least squares,M-PLS)建模方法,结合随机局部样本、局部选参、局部非连续性可调、交叉检验等技术相结合的方式建立局部优化模型,分别测试120个贮备饲料样本和120个青绿饲料样本中的氮(nitrogen,N)、中性洗涤纤维(neutral detergent fiber,NDF)、酸性洗涤纤维(acid detergent fiber,ADF)含量。将贮备饲料的定标校正模型应用于贮备饲料验证样本的营养成分测定,其标准误差(square error of prediction,SEP):N为1.02、NDF为16.56和ADF为13.47,相关系数均在0.9以上,相对预测偏差(relative prediction derivation,RPD)均大于3;该模型具有对青绿饲料样本的营养成分预测能力,其预测SEP:N为0.90、NDF为14.11和ADF为9.98,预测相关系数均在0.9以上,预测RPD均大于3,达到快速检测误差标准。由于局部建模过程中考虑了数据的潜在非线性结构和具有近似光谱响应的样本之间的不均匀性,相对全局建模方式而言具有更好的数据驱动性质,其建模效果优于全局建模方法。结果表明,基于贮备饲料样本建立的NIR定标校正模型可以用于青绿饲料营养成分的预测,特别是局部分析模型的应用能够提高NIR快速分析的预测精度。

Measuring nutrient content of green forage in silage using near-infrared spectroscopy

Abstract: Real-time monitoring of nutrient content of green forage in silage is essential to understanding how the nutrition change with time. In this paper we present a method to estimate nutrient content of the green forage using near-infrared (NIR) spectroscopy. Based on calibrated NIR models, a optimization model was modified and applied to estimate the nutrient. All green forage samples were collected from a grassland and their spectroscopy scanning was conducted in laboratory under controlled temperature and humidity. The results of 230 samples were used to train the chemometric algorithmic model, and the local optimization model was constructed using the modified partial least squares (M-PLS) algorithm combined with local random sample technique, local optimization and discontinuous adjustment of model parameters, and cross validation. For both silage and green forages, we measured nitrogen content, neutral detergent fiber (NDF) and acid detergent fiber (ADF) in 120 samples each. As a comparison, a global calibration model was also constructed based on the full-length waveband and applied to validate against the silage forage samples. The results showed that the square error of the prediction was 1.02 for nitrogen, 16.56 for nutrient NDF and 13.47 for nutrient ADF. The standard prediction errors were small and the correlation coefficients were higher than 0.9, with the relative derivation greater than 3. The model calibrated against the silage forage samples was able to predict nutrient content in both silage samples and green forage samples with SEP being 0.90 for nitrogen, 14.11 for NDF and 9.98 for ADF. The associated correlation coefficients were higher than 0.9, with the RPD greater than 3. All these results meet the standard for fast detection. The model calibrated locally can deal with non-linear molecular structure and non-uniform response of NIR spectroscopy. Experimental examination revealed that the locally calibrated model was more effective than the global model, and we can thus conclude that the NIR calibration model against the silage samples is able to predict nutrient content of green forage samples, especially the locally calibrated model.