基于卷积神经网络的阅海湖水质指标反演模型构建

Construction of Water Quality Index Inversion Model of Yuehai Lake Based on Convolutional Neural Network

水质检测对水生态污染治理和环境保护有重要意义。以银川市阅海湖湿地为例,基于Landsat-8影像数据和取样水质参数实测值,首先建立了输入节点数为8、卷积层为2、归一化层、全连接层为16-1和16-8-1、池化层和回归输出层均为1的卷积神经网络反演模型,对化学需氧量(COD)、氨氮(NH 3-N)、总磷(TP)、总氮(TN)4项水质参数浓度进行反演。其次,选用不同的卷积核、全连接层和优化器进行对比试验。结果表明,卷积模型14适用于COD等浓度值>3 mg/L的水质指标反演,决定系数R 2为0.91;卷积模型10适用于氨氮、总磷、总氮等浓度值范围在3 mg/L以内的水质指标反演,决定系数R^(2)分别为0.95、0.81、0.9;卷积反演模型的反演精度(R^(2),R RMSE)达到(0.91,4.44),优于BP神经网络和传统的水质反演方法。最后,对水质指标COD进行了数据可视化处理,制作卷积神经网络预测COD浓度值热图,结果表明在空间上阅海湖湿地COD污染程度中间水域高于南部和北部水域。

Water quality testing is of great significance to water ecological pollution control and environmental protection.Taking Yuehai Lake Wetland in Yinchuan City as the research object,based on the Landsat-8 image data and the measured values of sampling water quality parameters,a convolutional neural network inversion model was established which include 8 input nodes,2 convolutional layer,16-1 normalization layer,16-8-1 the fully connected layer,1 pooling layer and 1 regression output layer.The concentrations of four water quality parameters,such as chemical oxygen demand(COD),ammonia nitrogen(NH 3-N),total phosphorus(TP)and total nitrogen(TN),are inverted.Secondly,different convolution kernels,fully connected layers and optimizers were selected for comparative experiments.The results show that:the convolution model 14 is suitable for the inversion of water quality indexes with COD concentration values>3 mg/L,and the coefficient of determination R 2 is 0.91;the convolution model 10 is suitable for the inversion of water quality indicators with concentration values of ammonia nitrogen,total phosphorus and total nitrogen within 3 mg/L,and the coefficients of determination R 2 are 0.95,0.81,and 0.9,respectively;The inversion accuracy(R 2,R RMSE)of the convolution inversion model reached(0.91,4.44),which was better than the BP neural network and the traditional water quality inversion method.Finally,the data of COD was visualized,and a heat map of COD concentration value predicted by convolutional neural network was produced,and the results show that the middle waters of COD pollution in the wetlands of Yuehai Lake were higher than those in the south and north.