基于Vis-NIR光谱的土壤质地BP神经网络预测

Estimation of Soil Texture Based on Vis-NIR Spectroscopy and BP Neural Network

为快速、准确地获取土壤质地信息,提出了应用Vis-NIR光谱结合BP神经网络的建模方法。以河南封丘县的86个土壤样本为研究对象,以原始光谱和微分光谱主成分为输入变量,建立土壤粘粒和砂粒的BP神经网络预测模型,并将其预测结果与多元线性逐步回归模型进行比较。结果表明:基于原始光谱主成分的BP人工神经网络预测结果最好,优于多元逐步回归模型,预测粘粒和砂粒的RMSE分别为1.62和6.52。BP神经网络所建模型训练时间短、准确度也较高,能实现对土壤质地的高效预测。

In order to determinate the soil texture fast and accurately, this paper put forward modeling a method of Vis-NIR spectral analysis technology combined with BP neural network.Based on 86 soil samples collected from Fengqiu County, BP neural network models of clay content and sand content were established using PCA of the original and differential spectral, and compared with the correction model of MLSR. The result showed that BP neural network was better than MLSR ,Root-Mean-Square Error of Prediction （RMSE） of the model of clay content and sand content were 1.62 and 6.52 respectively. BP neural network model had short training time and high accuracy ,which was able to achieve rapid and efficient prediction on the soil texture.