基于深度稀疏学习的土壤近红外光谱分析预测模型

Soil Near-infrared Spectroscopy Prediction Model Based on Deep Sparse Learning

提出一种基于深度稀疏学习的土壤近红外光谱分析预测模型。首先,使用稀疏特征学习方法对土壤近红外光谱数据进行约简,实现土壤近红外光谱内容的稀疏表示;然后采用径向基函数神经网络以稀疏表示特征系数为输入,以所测土壤成分为输出,分别建立土壤有机质、速效磷、速效钾的非线性预测模型。结果表明用该模型预测土壤有机质的含量是可行的,但对土壤速效磷和速效钾含量的预测还需对模型做进一步的优化。

This paper presents a soil near-infrared spectroscopy prediction model based on sparse representation and radial basis function neural. The model first makes the soil near-infrared large spectroscopy data to be sparse,then the model uses radial basis function neural network with sparse representation coefficients as input and the measured soil composition value by chemical methods as output to establish effective nonlinear predictive model of soil organic matter,available phosphorus and potassium respectively. The results show that the model is feasible to predict soil organic matter content,but the model needs to be further optimized on the soil phosphorus or potassium effective prediction.