多光谱结合神经网络预测模型的总磷检测研究

Research of Total Phosphorous Detection Combined Multi-Spectra Spectroscopy with Neural Network Algorithm

多光谱遥感应用于水质参数检测具有良好的应用前景,但是目前基于多光谱的水质参数反演模型大多使用经验统计模型,通用性差。针对上述问题,以鄱阳湖为例,共采集了41组Sentinel-2A卫星的多光谱遥感数据,其中35组光谱数据作为训练集,建立了基于神经网络算法的水质总磷浓度预测模型,该模型内部验证和测试的相关系数0.8以上。剩余6组光谱数据作为外部测试集,预测值与真实值的相关系数为0.88,均方根误差为0.048。实验结果表明基于神经网络算法的多光谱总磷浓度检测模型是一种可行的水质遥感检测技术。

The application of multispectral remote sensing to water quality parameter detection has a good application prospect, but at present, most inversion models based on multispectral water quality parameters use an empirical statistical model, which has poor universality. To solve the above problems, 41 sentinel-2A satellite multi-spectral remote sensing data were collected from Poyang Lake, 35 of which were used as training sets. A prediction model of total phosphorus concentration in water quality based on the neural network algorithm was established. The correlation coefficient of internal verification and testing of the model was above 0.8. The re-maining six groups of spectral data were used as external test sets, and the correlation coeffi-cient between predicted value and true value was 0.88, and the root mean square error was 0.048. The experimental results show that the multi-spectral total phosphorus concentration detection model based on neural network algorithm is a feasible water quality remote sensing detection technology.