基于卷积神经网络的东海岛屿地下水分布遥感预测

Prediction of Groundwater Distribution of Islands in East China Sea Based on Convolutional Neural Network by Remote Sensing

为预测东海岛屿地区的地下水分布情况,研究以坡度、植被覆盖度、土壤湿度、地表温度等地下水关联的遥感指标作为岛屿地区地下水富集性评估的影响因子,进行地下水富集性综合评估,得出评估等级图。再根据地调及物探等资料对网格化的地下水富集性评估等级图进行贴标签工作,利用卷积神经网络对样本数据进行学习、训练与测试,得出岛屿地区地下水分布预测模型。以东海某岛屿作为研究区,以GF-1、DEM、Landsat8等为数据源,利用遥感技术进行地下水富集性评估,将评估结果图剖分为16×17的网格,每个小格约15万平方米,结合地调等资料制作样本集。以8∶2划分训练集与测试集,经训练,损失为0,精确度为100%,且测试的准确度达90%,得到了一个东海岛屿地下水水量分布遥感预测模型,该模型可预测该海域其他岛屿的地下水水量分布情况。

In order to predict the groundwater distribution of islands in East China Sea,this paper took 5 groundwater related indexes such as slope,vegetation coverage,soil humility,land temperature,etc.as the influencing factors for assessing groundwater enrichment in island areas.On this basis,a comprehensive assessment of groundwater enrichment was conducted and the rating map of the target area was obtained.Then,the map of the groundwater enrichment was gridded and labeled based on geological survey and geophysical data.The prediction model of groundwater distribution in the target area was obtained by learning,training and testing the sample data with convolutional neural network.In this paper,a certain island area in East China Sea was taken as the targeted sample,and GF-1,DEM,Landsat8 and other data sources were used to conduct groundwater enrichment assessment by remote sensing technology.The assessment results were divided into 16×17 grids with each grid about 150000 square meters and a sample set was made combined with the data of geological survey and other data.The sample set was divided into the training set and the test set by 8∶2.After training,the loss was 0,the accuracy was 100%,and the accuracy of the test was up to 90%.A prediction model of the groundwater distribution of islands in the East China Sea was obtained,which can be applied to predict the groundwater distribution of other islands in this sea area.