基于Kubelka – Munk理论的稻田土壤含水率高光谱反演研究

Hyperspectral Inversion of Soil Moisture Content in Paddy Fields Based on Kubelka – Munk Theory

摘要:

稻田土壤含水率是影响水稻生长的关键指标,利用高光谱技术实现其快速检测具有重要应用价值。针对高光谱在稻田土壤含水率宽幅变化条件下反演精度受限的问题,基于Kubelka – Munk(KM)理论建立辐射传输模型,利用实测稻田土壤高光谱及含水率逐波长求解吸收系数和散射系数,并通过2种途径提高含水率反演精度。以吸收散射比作为输入特征,采用最小二乘(Ordinary least squares, OLS)与多层感知机(Multi-layer perceptron, MLP)结合构建含水率半经验模型(KM – OLS – MLP);利用吸收系数和散射系数重建模拟光谱,扩充训练集样本实现数据增强(KM-data augmentation,KM – DA),并构建含水率预测模型。结果表明,稻田土壤质量含水率在26. 33% ~61. 09%区间时,KM – OLS – MLP模型精度最高,测试集决定系数R2为0. 924 1,均方根误差RMSE为2. 41% ;模拟光谱在测试集上的RMSE和结构相似性指数(Structural similarity index, SSIM)分别为0. 008 1和0. 996 4。数据增强后,偏最小二乘回归(Partial least squares regression, PLSR)、支持向量回归(Support vector regression, SVR)以及BP神经网络(Back propagation neural network, BPNN)模型精度均得到提高。BPNN模型精度提高最显著,模拟光谱与实测光谱比例为2∶1时,测试集R2从0. 790 6提高至0. 898 7,RMSE从4. 00%降低至2. 78% 。该研究可为含水率宽幅变化条件下的稻田土壤含水率高光谱反演提供参考。

Abstract:

Soil moisture content in paddy fields is a critical indicator affecting rice growth, and its rapid detection using hyperspectral technology holds significant application value. To address the limitations of hyperspectral inversion accuracy under wide-ranging variations in paddy soil moisture content, a radiative transfer model was established based on the Kubelka – Munk ( KM) theory. Using the measured hyperspectral data and the corresponding soil moisture content values of paddy soil, the absorption and scattering coefficients were estimated at each wavelength. Two approaches were employed to enhance the inversion accuracy of soil moisture content. Firstly, a semi-empirical model ( KM – OLS – MLP) was developed by using the absorption-to-scattering ratio as an input feature, combining ordinary least squares (OLS) and multi-layer perceptron ( MLP). Secondly, simulated spectra were reconstructed using the absorption and scattering coefficients to augment the training dataset ( KM-data augmentation, KM – DA), and a soil moisture prediction model was constructed. The results showed that when the mass moisture content of paddy soil ranged from 26. 33% to 61. 09% , the KM – OLS – MLP model achieved the highest accuracy, with a coefficient of determination (R2) of 0. 9241 and a root mean square error (RMSE) of 2. 41% on the test set. The simulated spectra exhibited an RMSE of 0. 008 1 and a structural similarity index (SSIM) of 0. 996 4 on the test set. After data augmentation, the accuracy of the partial least squares regression ( PLSR), support vector regression ( SVR), and back propagation neural network (BPNN) models was improved, with BPNN model showing the most significant enhancement. When the ratio of simulated to measured spectra was 2∶1, the test set R2 was increased from 0. 790 6 to 0. 898 7, and the RMSE was decreased from 4. 00% to 2. 78% . The research result can provide a valuable reference for hyperspectral inversion of soil moisture content in paddy fields under wide-varying moisture conditions.