极限学习机模型的土壤含水量反演研究

Soil water content inversion based on extreme learning machine model

针对传统反馈型神经网络模型在土壤含水量反演时容易陷入局部最优解和模型计算效率低等问题,该文提出了极限学习机模型结合主被动遥感进行土壤含水量反演的方法。首先,使用水云模型计算裸土后向散射系数,通过高级积分方程模型建立组合地表粗糙度库,计算各点的地表粗糙度;其次,以计算的裸土后向散射系数、植被指数、地表粗糙度和入射角作为输入数据,以土壤含水量为输出,构建极限学习机模型,并进行训练;最后,对极限学习机土壤含水量反演结果进行验证。结果表明,该方法反演土壤含水量具有较高的精度和计算效率;同时,与BP神经网络模型的反演结果比较,证明了该方法的有效性,为土壤含水量的反演研究提供了一种方法。

According to the fact that traditional feedback neural network models tend to fall into local optimal solutions and the low calculation efficiency of the model when inverting soil water content,an inversion method for soil water content was presented by using an extreme learning machine model combined with active and passive remote sensing.Firstly,the water cloud model was used to calculate the backscattering coefficient of bare soil,and the combined surface roughness library was established through the advanced integral equation model to calculate the surface roughness of each point.Secondly,the calculated backscattering coefficient,vegetation index,surface roughness and incident angle were used as input data,and soil water content was used as output data to build an extreme learning machine model and conduct training.Finally,the soil water content inversion results of extreme learning machine were verified.The results indicated that this method had high accuracy and efficiency in soil water content retrieval.At the same time,compared with the inversion results of BP neural network model,the effectiveness of this method was proved,which provided a method for the inversion research of soil water content.