利用卷积神经网络的土壤有机质含量高光谱估测

Hyperspectral estimation of soil organic matter content using convolutional neural network

为提高土壤有机质高光谱估测精度,以山东省济南市章丘区的76个土壤样本有机质含量及其高光谱数据为基础,建立基于卷积神经网络的土壤有机质含量估测模型。首先对原始高光谱数据进行预处理,利用主成分分析对光谱数据降维,并转化为四维光谱信息数组,通过实验模拟调整各项参数及网络结构得到最优估测模型。结果表明:当模型采用1个3×3的卷积核,1个平均池化层,1个完全连接层,且网络计算迭代600次时,卷积神经网络模型达到最优预测效果,其中12个检验样本估测结果的决定系数为R^(2)=0.841,平均相对误差为7.123%,精度均优于传统模型。研究表明利用卷积神经网络估测土壤有机质含量是可行有效的。

In order to improve the accuracy of hyperspectral estimation of soil organic matter,this paper built an estimation model of soil organic matter content using the convolutional neural network and hyperspectral data of 76 soil samples collected from Zhangqiu district,JINAN city,Shandong Province.Firstly,the original hyperspectral data was preprocessed,and principal component analysis was used to reduce the dimension of the hyperspectral data whose results was transformed into a four-dimensional spectral information array.Through experimental simulation,various parameters and network structure were adjusted to obtain the optimal estimation model.The results showed that the convolutional neural network model achieves the optimal prediction effect when the model adopted a 3×3 convolution kernel,an average pooling layer,a fully connected layer and 600 times iteration.The R^(2) of the 12 test samples was 0.841,and the mean relative error was 7.123%.The accuracy was better than that of the traditional models.Our results suggested that it is feasible and effective to estimate soil organic matter content using convolutional neural networks.