基于反射光谱的江淮分水岭区域典型农作物识别

对江淮分水岭区域观测的8种农作物冠层光谱进行数据重采样和植被指数计算,分析了4种常用指数和6种常用传感器对农作物的识别能力,同时采用识别效率最高的数据变换形式构建了BP神经网络模型。结果表明:8种农作物的反射光谱曲线存在较大差异;6种传感器对农作物的识别能力由大到小依次为ETM+、QUICKBIRD、IKONOS、MODIS、ASTER、HRG;模拟得到的ETM+和QUICKBIRD的近红外与红光波段反射率计算的归一化植被指数(NDVI)和简单比值植被指数(SR)对农作物的识别能力较强;在不同的数据变换形式中,对农作物识别精度最高的是一阶微分(FD,波长间隔6 nm),识别精度达87.3%;以FD(波长间隔6 nm)为输入数据集构建BP神经网络模型,当隐含层节点数为15时,识别精度最高,达90.0%。

Typical crop species identification based on the spectral reflectance in Jianghuai watershed area

Data resample and vegetation index calculation were used to deal with the observed eight kinds of crop canopy spectra in Jianghuai watershed area, and the crop species recognition ability for four common indexes and six sensors was analyzed. At the same time, the data transformation form with the highest recognition efficiency was used to construct the BP neural network model. The results showed that eight kinds of crop spectral curves had large differences and the recognition ability of 6 sensors from big to small: ETM +, QUICKBIRD, IKONOS, MODIS, ASTER, HRG. Crops recognition ability of normalized difference vegetation index (NDVI) and simple ratio (SR) computed by near infrared and red band reflectance of ETM + and QUICKBIRD was stronger. First order differential (FD) (wavelength interval 6 nm) had the highest identification accuracy among different data transformation forms and the identification accuracy was 87.3%. The BP neural network model with 15 hidden layer nodes built by FD (wavelength interval 6 nm) had the highest recognition accuracy, up to 90.0%.