以表层土壤为对象, 探究土壤的多光谱反射率与土壤水分含量相关性, 进行基于无人机多光谱图像的土壤水分含量预测模型方法的探究。 选取中国农业大学通州实验站为研究区域, 实地采集试验田的土壤样本100组, 按照一定梯度配制土壤含水量, 配成的土壤含水率为10%~50%之间, 土壤含量的真实值采用土壤烘干法进行测定。 多光谱相机灵巧便捷, 可搭载在无人机上对土壤进行监测。 将RedEdged-M型多光谱相机搭载在Phantom 3型无人机上, 选择阳光充足的采集环境, 实时采集土壤样本的多光谱图像, 建立土壤多光谱信息与水分含量之间的模型。 利用处理光谱数据的ENVI5.3软件提取土壤样本多光谱信息, 以多光谱相机自带的标准白板反射率为100%, 计算出土壤样本在蓝、 绿、 红、 红边、 近红外五个波段的光谱反射率。 采用BP神经网络算法、 支持向量机算法、 偏最小二乘算法分别建立基于无人机多光谱图像的土壤水分含量的预测模型。 以80组土壤样本数据作为训练集, 建立基于多光谱图像的土壤水分含量预测模型。 采用莱文贝格-马夸特算法对BPNN进行改进, 提高了其训练速度, 当网络结构为5-10-1时, 训练效果最好, 本文选择该网络结构; SVM采取高斯核函数, 当参数为0.56时, 模型效果最好。 本研究采用归一化均方根误差(NRMSE)和决策系数(R2)对三种土壤水分含量的预测模型进行定量对比。 以20组土壤样本数据作为测试集, 结果可知, 基于BP神经网络土壤水分含量预测模型的NRMSE为0.268, R2为0.872; 基于支持向量机的土壤水分含量预测模型的NRMSE为0.298, R2为0.821; 基于偏最小二乘土壤水分含量预测模型的NRMSE为0.316, R2为0.789。 对三种模型分析可知, 基于BPNN的土壤水分含量预测模型效果均较好。 结果可知, 土壤的光谱反射率与含水率间存在较密切的相关性, 将多光谱相机搭载在无人机上可以对土壤水分含量进行有效的实时监测, 对监测土壤墒情提供技术支持和理论支撑。

In this paper, we studied the correlation between multispectral soil reflectance and soil moisture content. In this paper, the Tongzhou experimental station of China Agricultural University was selected as the research area. In the experimental wheat field of this experimental station,wecollected100 groups of soil samples on the spot. According to a certain gradient, we prepared the soil moisture content between 10% and 50%. The real value of soil content was determined by drying method. Multispectral cameras are convenient. Red Edged-M multispectral camera was mounted on innovative Phantom 3 UAV. The multispectral images of soil samples were collected in a sunny environment. The multispectral images of soil samples were processed. The spectral reflectance of soil samples was extracted, and the model between multispectral soilinformation and water content was established. The multispectral information of soil samples was extracted by ENVI5.3 software.We, asthe reflectance of the standard whiteboard is 100%. The spectral reflectance of soil samples in five bands of blue, green, red and near-infrared was calculated. In order to explore the correlation between spectral reflectance and moisture content of soil samples, we established BP neural network algorithm, support vector machine algorithm, andpartial least squares algorithm prediction models of soil moisture content based on UAV multispectral images. Based on 80 sets of soil sample data as a training set, a prediction model of soil moisture content based on the multi-spectral image was established. The BP neural network algorithm is improved bythe Levenberg-Marquardt method, which improves the training speed of its BP neural network model. When the network structure is 5-10-1, the number of iterations is the least, and MSE is the least. This paper chooses the network structure. The support vector machine algorithm adopts the Gauss kernel function, and when the parameter is 0.56, the model has the best effect. In this study, normalized root means square error (NRMSE) and decision coefficient were used to compare the three prediction models of soil moisture content quantitatively. We used 20 sets of soil sample data as the test set, the results showed that the normalized root mean square error of soil moisture content prediction model based on BP neural network was 0.268, and the decision coefficient was 0.872; the normalized root mean square error of soil moisture content prediction model based on support vector machine was 0.298, and the decision coefficient was 0.821; The normalized root mean square error (NRMSE) of the prediction model multiplied by soil moisture content is 0.316, and the decision coefficient is 0.789. According to the analysis of the three models, the prediction model of soil moisture content based on BP neural network has a good effect. Through the research, we can know that itis a close correlation between soil spectral reflectance and water content. The multi-spectral camera can effectively monitor soil moisture content in real-time on the UAV. In this study, multi-spectral technology is applied in the field of soil moisture content detection to provide technical support and theoretical support for monitoring soil moisture.