基于遗传神经网络的克钦湖叶绿素反演研究

叶绿素a浓度能够在一定程度上反映内陆湖泊水质情况。为实现对克钦湖水体叶绿素a浓度的监测,于2010年8月15日对克钦湖进行了现场光谱测量和同步采样。通过分析叶绿素a浓度和光谱数据之间的关系,建立基于反射比、人工神经网络和遗传神经网络的叶绿素a浓度估测模型。结果表明:利用R700nm/R670nm反射比建立的模型估测精度为R2=0.67;人工神经网络模型的估测精度较高,R2=0.882;将遗传算法引入神经网络之后,模型的估测精度进一步提高,R2达到0.956,将模型预测的结果与克里格内插法相结合对研究区的叶绿素a空间分布情况进行定量估测,发现北湖的叶绿素a浓度明显高于南湖,有由北向南逐渐递减的趋势,这为今后利用高光谱数据对克钦湖叶绿素a浓度大面积遥感反演提供了研究基础。

Research on Chlorophyll Concentration Retrieval Models of Keqin Lake based on Genetic Neural Networks

The concentration of Chlorophyll-a could reflect the water quality of inland lakes to some extent.In order to monitor the concentration of chlorophyll-a,hyperspectral reflectance was measured from July to August in 2010 with ASD FieldSpec HH in Keqin Lake.Concurrently,water samples were collected.Three models including spectral ratio,artificial neural networks,and genetic neural networks were developed by analyzing the correlations between concentration and hyperspectral reflectance data.The results show that spectral ratio gives determination coefficient R2 = 0.67.Artificial neural networks gives better results with higher determination coefficient R2 = 0.882 which was further improved to 0.956 after introducing genetic algorithm to neural networks.All of the three models with significance level P<0.01,and are applied to estimate chlorophyll-a concentration.Finally,the author used the predicted results of the GANN model and the Kriging analysis technique to obtain the quantitative estimation of the spatial distribution of Chlorophyll-a in the study area.These algorithms provided a research basis of future large area remote sensing inversion with hyperspectral data in Keqin Lake.