倒伏水稻特征分析及其多光谱遥感提取方法研究

倒伏水稻的识别对灾后农业生产管理、灾害保险、补贴等工作有重要意义。为应用高分辨率遥感影像准确提取倒伏水稻面积,本文利用2019年9月27日获取的哨兵2号多光谱遥感影像,研究黑龙江省同江市倒伏水稻的光谱、纹理特征,并基于光谱与纹理特征建立倒伏水稻的遥感提取模型。研究结果表明水稻倒伏后可见光-近红外-短波红外等8个波段的反射率均升高,其中短波红外、红光和红边1等3个波段的反射率上升大于0.06。倒伏水稻的典型植被指数中,归一化植被指数、比值植被指数、增强植被指数和红边位置指数均降低,但差值植被指数升高。倒伏与正常水稻在红光、红边1和短波红外等3个波段的均值纹理数值差距明显,红光波段的纹理均值差异最大。利用归一化植被指数、地表水分指数、比值植被指数和差值植被指数以及红光波段的纹理均值构建决策树分类模型,监测结果表明农场内倒伏水稻分布较散,其西部和南部水稻受灾面积较大,北部受灾面积较小,中部偏北和东部基本未倒伏。将本文模型所提取的结果与实测面积对比,正常与倒伏水稻的面积识别误差分别为3.33%和2.23%。利用随机验证样本与模型验证结果进行混淆矩阵分析,倒伏水稻的用户精度和制图精度均为92.0%,Kappa系数为0.93。该方法能够适用于大区域倒伏水稻提取,可为高分辨率多光谱遥感数据调查水稻倒伏面积提供相关依据。

Characteristic analysis of lodging rice and study of the multi-spectral remote sensing extraction method

Crop lodging assessment is essential for evaluating yield damage and informing crop management decisions for sustainable agricultural production. Traditional evaluation methods and manual on-site measurements are time-consuming and labor- and capital-intensive. In this study, a remote sensing model to distinguish lodging rice was constructed based on spectral and textural features. To accurately extract the area of lodging rice from high-resolution remote sensing images, this study used Sentinel-2 multispectral images taken on September 27, 2019, to study the spectral and textural characteristics of lodging rice, in Tongjiang City, Heilongjiang Province. Analysis of the surface reflectance of normal rice and lodged rice, showed that reflectance of eight bands, including visible light, near-infrared, and shortwave infrared, increased after rice lodging; the reflectance of shortwave infrared, red light, and red edge 1 increased by more than 0.06. Except for the difference vegetation index (DVI), the typical vegetation indices of lodged rice, such as normalized difference vegetation index (NDVI), ratio vegetation index (RVI), enhanced vegetation index (EVI), and red edge position index (REP), decreased. There were significant differences between lodging rice and normal rice in the mean texture feature values of the red band, red edge 1, and shortwave infrared; the largest difference was for the mean texture value of the red band. Therefore, in this study, normalized difference vegetation index, land surface water index (LSWI), ratio vegetation index, difference vegetation index, and texture mean of the red band were used to construct the decision tree classification model. The results of remote sensing monitoring showed that rice lodging on the farm was decentralized. The area of rice disaster was larger in the west and south and smaller in the north. There was no lodging rice in the middle of the north and the east. Compared with the measured area, the area recognition errors of normal and lodged rice were 3.33% and 2.23%, respectively. When using random verification samples and model verification results for the confusion matrix analysis, the user accuracy and mapping accuracy of lodging rice were 92.0%, and the Kappa coefficient was 0.93. These results show that this method can be applied to remote sensing data from lodged rice in large areas and can provide a relevant basis for the investigation of rice lodging areas using high-resolution and multi-spectral remote sensing data.