基于超宽带雷达和多光谱数据融合的土壤含水率检测

土壤含水率监测是精准农业的重要组成部分,对于农情监测和农业生产起着关键性作用。超宽带雷达由于其体积小、质量轻、穿透力强和功耗低等特性已被广泛应用于土壤含水率监测研究。而现有超宽带雷达反演土壤含水率多为理想裸土情况,实际应用中地表植被覆盖会对结果造成较大影响,针对此问题,融合超宽带雷达和多光谱数据,利用支持向量机（SVM）模型对农田尺度不同植被覆盖下的土壤含水率进行分级预测,以减小植被对预测精度的影响。研究结果表明,在超宽带雷达回波数据提取出的不同时域特征组合中,选用峰值因子、峭度、均方根、峰-峰值、最大幅值、方差、偏斜度、平均值和最小幅值9个时域特征作为SVM模型输入特征预测结果最好,总体精度为95.59%,Kappa系数为0.9492。加入植被指数NDVI后,不同时域特征组合作为特征输入的模型精度均有显著提高,其中将9个时域特征与NDVI共同作为SVM输入预测效果最佳,总体精度为98.09%,Kappa系数为0.9780,与不考虑植被影响的预测结果比较,总体精度提高了2.50个百分点,Kappa系数提高了0.0288。

Monitoring Method of Soil Moisture Based on Ultra-wide Band Radar and Multispectral Data

Soil moisture monitoring is an important part of precision agriculture, and it plays a key role in monitoring agricultural conditions and agricultural production. Ultra-wide band (UWB) radar has been widely used in soil moisture monitoring due to its small size, light weight, strong penetrating power and low power consumption. However, most of soil moisture retrieved with UWB radar is for the case of ideal bare soil conditions. In practical applications, surface vegetation coverage will have a great impact on the results. To solve this problem, support vector machines (SVM) model was used to predict soil moisture under different vegetation coverages in farmland scale by combining UWB radar and multispectral data, so as to eliminate or mitigate the effect of vegetation coverage. The experimental results showed that among different time-domain feature combinations extracted from UWB radar echo data, SVM model with the inputs of nine selected time domain features, including crest factor, kurtosis, root mean square, peak-to-peak value, maximum amplitude, variance, skewness, average and minimum, generated the best prediction results, and the overall accuracy and Kappa coefficient were 95.59% and 0.9492, respectively. After adding the normalized difference vegetation index (NDVI), the model accuracy for different time-domain feature combinations was significantly improved. Among them, the results by combining the nine selected time-domain features and NDVI were the best, and the overall accuracy and Kappa coefficient reached 98.09% and 0.9780, which were raised by 2.50 percentage points and 0.0288 compared with those without considering the influence of vegetation covers.