| 引用本文: | 蔡建楠,刘海龙,姜波,等.基于GA-PLS算法的河网水体化学需氧量质量浓度高光谱反演[J].灌溉排水学报,2020,():-. |
| CAI Jiannan,LIU Hailong,JIANG BO,et al.基于GA-PLS算法的河网水体化学需氧量质量浓度高光谱反演[J].灌溉排水学报,2020,():-. |
|
| 本文已被：浏览 1822次 下载 0次 |
|
|
|
| 基于GA-PLS算法的河网水体化学需氧量质量浓度高光谱反演 |
|
蔡建楠1, 刘海龙2, 姜波2, 何甜辉3, 陈文杰4, 冯志伟4, 黎倬琳4, 邢前国2
|
|
1.中山市环境监测站 2.中山市生态环境局;2.中国科学院烟台海岸带研究所/中国科学院海岸带环境过程与生态修复重点实验室;3.中山市环境监测站;4.中山市生态环境局
|
|
| 摘要: |
| 【目的】为了建立河网水体化学需氧量（COD）质量浓度高光谱反演模型，验证遗传-偏最小二乘（GA-PLS）算法对建模效果的改善作用。【方法】采集广东省中山市146个点位的水体高光谱数据和COD质量浓度实测数据，通过GA-PLS算法对高光谱反射率数据进行特征波段筛选后建立COD质量浓度反演模型，并比较输入变量为不同特征波段组合时模型反演效果差异。【结果】基于GA-PLS算法的COD质量浓度高光谱模型反演效果优于全谱段PLS模型，验证集RMSEP最小为4.887 mg/L，较全谱段PLS模型降低11.4 %；以筛选得到的74 个波段（占全波段数的2.9 %）作为输入变量时，模型仍可保持良好的稳定性和反演精度；GA-PLS算法筛选得出的部分特征波段与水体中藻类、悬浮颗粒物的吸收特征波段相一致，筛选结果具有合理性和指示意义。【结论】通过GA-PLS算法可对高光谱数据进行特征波段筛选，实现数据降维优化，进一步简化模型；在样本COD质量浓度主要分布范围内，GA-PLS算法模型有良好的反演精度和水质类别分类准确性。该方法在河流COD快速监测中具有良好的应用前景。 |
| 关键词: 高光谱；遗传算法；偏最小二乘法；化学需氧量；河网水体 |
| DOI： |
| 分类号:X832 |
| 基金项目:中国科学院重点仪器项目(No. YJKYYQ20170048)、国家自然科学项目(No.41676171)、2020年广东省科技创新战略专项(No. PDJH2020b1091) |
|
| Hyperspectral Retrieval of Chemical Oxygen Demand for River Network Waters Based on GA-PLS Algorithm |
|
CAI Jiannan1, LIU Hailong2, JIANG BO2, HE Tianhui3, CHEN Wenjie4, FENG Zhiwei4, LI Zhuolin4, XING Qianguo2
|
|
1. Zhongshan Environmental Monitoring Station 2. Zhongshan Ecology and Environmental Agency;2.Key Laboratory of Coastal Environmental Processes and Ecological Remediation, Yantai Institute of Coastal Zone Research, Chinese Academy of Sciences;3.Zhongshan Environmental Montoring Station;4.Zhongshan Ecology and Environmental Agency
|
| Abstract: |
| 【Objective】Hyperspectral remote sensing has been proven to be an effective tool for water quality monitoring. However, there are also defects such as data redundancy and susceptibility to environmental noise, which affect the accuracy and stability of water quality retrieval by hyperspectral technique. The Genetic Algorithm-Partial Least Squares (GA-PLS) algorithm with the function of selecting sensitive spectral variables can overcome the above disadvantages. In the field of water quality remote sensing monitoring, GA-PLS was mainly used for the retrieval studies of optically active parameters such as transparency, chlorophyll-a, suspended matter and turbidity in lakes and reservoirs, but there were few reports on hyperspectral retrieval for comprehensive indicators of river water quality. The purpose of this paper is to establish a hyperspectral retrieval model for chemical oxygen demand (COD) concentration in the river network of the Pearl River Estuary and validate the improvement of GA-PLS algorithm on the modeling effect.【Method】Hyperspectral data and COD concentration data of 146 samples from the water body of the Pearl River Estuary were collected, and the characteristic bands of hyperspectral reflectance data were screened by GA-PLS algorithm to retrieve COD concentration. Differences in model retrieval effects with different band combinations as input variables were compared.【Result】The COD concentration hyperspectral retrieval model based on GA-PLS algorithm is better than the full-spectrum PLS model. The minimum RMSEP of the validation set is 4.887 mg·L-1, which is 11.4% lower than the full-spectrum PLS model. With 74 filtered bands (accounting for 2.9% of full bands) as input variables, the model can still maintain good stability and retrieval accuracy. Some of the characteristic bands obtained by the GA-PLS algorithm have clear physical meaning, indicating that the screening results are reasonable.【Conclusion】The GA-PLS algorithm can be used for characteristic bands selection of the hyperspectral data to optimize the dimension and further simplify the model. Within the main distribution range of the COD concentration of samples, the GA-PLS algorithm model has good accuracy for water quality retrieval and category classification accuracy. |
| Key words: hyperspectra; genetic algorithms; partial least squares; chemical oxygen demand; river network waters |
