近-中红外光谱融合技术速测奶牛场粪水氮磷含量

快速准确测定粪水/沼液中的氮磷养分含量，是现阶段我国规模化奶牛场在种养结合道路上亟需破解的技术瓶颈。针对传统的实验室湿化学检测方法难以满足奶牛场粪水还田前任一节点氮磷快速定量的现实问题，研究开发了基于近红外、中红外及近-中红外光谱融合技术，实现规模化奶牛场粪水运移全链条环节氮磷含量的本土化快速检测方法。采集了天津地区27家种养结合型规模化奶牛场粪水流经全程环节（集粪沟、集污池、氧化塘等）共计144个样品，使用傅里叶变换近红外光谱仪和中红外光谱仪分别采集了12 000～4 000和4000～650 cm-1区间的光谱数据，对所有样品的近红外光谱、中红外光谱及近-中红外光谱融合数据进行归一化、基线校正、SNV等预处理，分析了近红外光谱和中红外光谱特征，采用浓度梯度法进行样品分集，运用偏最小二乘（PLS）、间隔偏最小二乘（IPLS）和联合区间偏最小二乘（SIPLS）法，构建了粪水总氮（TN）、总磷（TP）的近红外模型和中红外模型。粪水TN模型预测结果较好，近红外和中红外最佳模型趋近一致；粪水TP模型的预测性能不理想，近红外和中红外SIPLS最佳模型的R2_pred仅分别为0.790和0.631，RPD分别为2.213和1.479，四分位数间隔（RPIQ）分别为3.616和2.351，难以用于实际检测；为实现粪水氮磷同步有效的测定分析，进一步提升模型整体预测性能，综合近红外光谱和中红外光谱数据建立了粪水氮磷近-中红外融合模型，光谱区间为12 000～650 cm-1，预测性能整体表现良好，以近-中红外融合IPLS模型预测结果最为理想，其R2_pred分别为0.970和0.861，RPD分别为5.615和2.684，RPIQ分别为12.874和4.394，总体优于单一近、中红外模型，尤其TP的最佳融合模型，其R2_pred相比近、中红外最佳单一模型分别高出0.071和0.170。研究表明，近-中红外光谱融合技术可以实现对规模化奶牛场粪水运移全链条环节氮磷含量的准确速测，为粪水科学还田提供技术支撑。

Rapid Determination of Nitrogen and Phosphorus in Dairy Farm Slurry Via Near-Mid Infrared Fusion Spectroscopy Technology

Rapid and accurate determination of the nitrogen (N) and phosphorus (P) in the slurry/biogas slurry has been a technical bottleneck is urgently needed for the large-scale dairy farms in China on their ways of planting and breeding combination. Conventional wet chemical measuring methods in the laboratory were difficult to meet the practical demand of rapid quantitative determination on the N and P before recycling the dairy farm slurry back to the field. An indigenized rapid detection method of N and P through the full chain of slurry movement in large-scale dairy farms was developed based on the near-infrared (NIR), mid-infrared (MIR) and near-mid infrared (NIR-MIR) spectral fusion technology. A total of 144 slurry samples were collected along with the entire process links (manure collecting gutter, slurry tank, lagoon, etc.) from 27 large-scale dairy farms in Tianjin. The spectral data of 12 000～4 000 and 4 000～650 cm-1 were collected by the Fourier transform near-infrared spectrometer (FT-NIRS) and mid-infrared spectrometer (FT-MIRS). Pretreatment methods involved the normalization, baseline and SNV were performed on the whole NIR, MIR and NIR-MIR data.NIR and MIR spectral characteristics were analyzed. The concentration gradient method was used for the sample diversity. NIR and MIR models of the total nitrogen (TN) and total phosphorus (TP) in the slurry were constructed by the partial least squares (PLS), interval partial least squares (IPLS) and synergy interval partial least squares (SIPLS). The results of slurry TN models were preferable, while the optimal models between the NIR and MIR were equivalent. The prediction performance of the TP model for the slurry was unsatisfactory that difficult of practical application. The R2_pred of the optimal SIPLS models for NIR and MIR were only 0.790 and 0.631, respectively. The residual predictive deviation (RPD) was 2.213 and 1.479 respectively. And the ratio of performance to interquartile range (RPIQ) was 3.616 and 2.351, respectively. In order to realize the simultaneous and effective determination and analysis of the N and P in the slurry meanwhile further improve the overall prediction performance of the model, the NIR-MIR fusion model of the N and P in the slurry was established integrated the NIR with MIR spectral data, with the spectral range of 12 000～650 cm-1. The prediction performance behaved well overall. IPLS fusion model performed the optimum. The R2_pred was 0.970 and 0.861 respectively. RPD was 5.615 and 2.684 respectively. RPIQ was 12.874 and 4.394 respectively. It was better than the single NIR model and MIR model. In particular, the optimal fusion model of the TP was 0.071 and 0.170 which was higher than that of the single NIR and MIR models. The results showed that exact and rapid determination of the N and P through the full chain links of slurry movement in large-scale dairy farms via the near-mid infrared spectroscopy fusion technology could be available for the scientific slurry recycling to the farmland.