水是维系生命与健康的基本需求, 人类的生产、 生活都离不开水。 水体中氮磷的超标导致水中营养元素过剩从而水体富营养化, 进而水质恶化甚至造成大范围的影响。 高光谱遥感在内陆水质监测领域的应用越来越广泛, 研究以此为基础, 为减少室外水体特异性因素影响, 通过实验室模拟外界条件搭建实验室实验系统, 并根据国家排放标准分别配制浓度范围在0～2.5 mg·L-1 的40个不同浓度梯度的磷酸钠标准溶液和浓度范围在0～20 mg·L-1 的40个不同浓度梯度氯化铵标准溶液。 获取所有标准溶液的高光谱图像, 对水质参数总磷、 总氮的光谱响应进行了分析, 找出其对应的敏感波段分别在420、 720 nm附近和410 nm附近。 利用主成分分析(PCA)建立高光谱水质反演数据集, 对高光谱图像作辐射定标、 Savitzky-Golay滤波(SG滤波)预处理并利用BP人工神经网络分别构建实验室高光谱总磷、 总氮反演模型, 构建的实验室高光谱总磷反演模型的决定系数为0.980 2, 实验室高光谱总氮反演模型的决定系数为0.860 2。 以江苏宜兴市内某河道为研究对象, 将该模型应用到室外无人机搭载高光谱成像系统获取到的室外高光谱图像数据, 分散选取五个点分别计算结果, 得到总磷、 总氮浓度均值的反演精度分别为95.00%和93.52%。 利用传统方法直接在待测河道观测点取水构建的室外高光谱水质反演模型反演相同五个点得到的总磷、 总氮浓度均值的反演精度分别为86.87%和86.48%。 两组反演结果对比, 发现本研究构建的实验室高光谱水质反演模型得到的光谱反演结果中90%的反演精度略高于室外水质反演模型的反演精度, 证实该研究可对待测河道内总磷、 总氮的含量进行有效预测, 也可为水体总磷、 总氮高光谱遥感反演提供一定技术支持。

Water is a basic need for life and health. Human production and life are inseparable from water. Excessive nitrogen and phosphorus in the water body lead to excess nutrients, resulting in eutrophication of the water body, and then the deterioration of water quality has a wide-ranging impact. The application of hyperspectral remote sensing in the inland water quality monitoring field is becoming more and more extensive. Based on this research, to reduce the influence of outdoor water body-specific factors, this study builds an experimental laboratory system by simulating external conditions in the laboratory. According to national emission standards, this study prepares 40 sodium phosphate standard solutions with different concentration gradients in the concentration range of 0～2.5 mg·L-1 and 40 different concentration gradient ammonium chloride standard solutions in the concentration range of 0～20 mg·L-1. After obtaining hyperspectral images of all standard solutions, The spectral responses of water quality parameters total phosphorus and total nitrogen were analyzed. This study finds the sensitive bands corresponding to total phosphorus and total nitrogen at around 420, 720 nm and around 410 nm. Building a hyperspectral water quality inversion dataset using Principal Component Analysis (PCA). By preprocessing hyperspectral image radiometric calibration, Savitzky-Golay filtering (SG filtering), and using the BP artificial neural network method to construct a laboratory hyperspectral water quality inversion model. The coefficient of determination of the constructed laboratory hyperspectral total phosphorus inversion model is 0.980 2, and the determination coefficient of the laboratory hyperspectral total nitrogen inversion model is 0.860 2. Taking an indoor river in Yixing, Jiangsu as the research object. The model is applied to the outdoor hyperspectral image data obtained by the outdoor UAV equipped with the hyperspectral imaging system. The inversion accuracies of the mean concentrations of total phosphorus and total nitrogen are 95.00% and 93.52%, respectively. The outdoor hyperspectral water quality inversion model constructed by using the traditional method to directly draw water from the observation points of the river to be tested the average values of total phosphorus and total nitrogen concentrations obtained at the same five points with an inversion accuracy of 86.87% and 86.48%. This study compares the inversion results of the two groups. It is found that the inversion accuracy of 90% of the spectral inversion results obtained by the laboratory hyperspectral water quality inversion model constructed in this study is slightly higher than that of the outdoor water quality inversion model. It is confirmed that this study can effectively predict the content of total phosphorus and total nitrogen in the river to be measured, and can also provide certain technical support for the hyperspectral remote sensing inversion of total phosphorus and total nitrogen in the water.