首页 > , Vol. , Issue () : 1993-2002
Aerial hyperspectral remote sensing dataset play an important role on the research of hyperspectral image, such as classification. However, few works has been done on the establishment of standard hyperspectral dataset. This paper introduced a standard hyperspectral dataset, which including hyperspectral remote sensing image, land cover map, and sensor parameters, the dataset was acquired by a newly designed airborne hyperspectral sensor, accompanied with the synchronous ground survey experiments. （1）Data acquisition：Aerial hyperspectral remote sensing image of Xiongan New Area was acquired using the visible and near infrared imaging spectrometer designed by Shanghai Institute of Technical Physics, CAS in October 2017. The total field of view angle of the spectrometer is 40.6 degree, the instantaneous field of view is 0.25 mrad, effective push-scan pixel is 2834, and the maximum speed-to-height ratio is 0.04. The flight height is 2000 m, and the flight areas cover the Xiong County, An County, Rong County and Baiyangdian Lake, of which the East-West length is 48 km, the North-South width is 27.5 km, and the total area is 1320 km2. There are 21 flight lines with east-west direction, while the Matiwan Village is located in the 10th and 11th flight line. The flying weather is clear and cloudless, and the visibility condition is good. And radiation correction, geometric correction, image mosaic and clipping were done before data classification. （2）Dataset：The spectral range of the aerial hyperspectral remote sensing image of Xiongan New Area (Matiwan Village) is 400-1000 nm, with 250 bands and a spatial resolution of 0.5m. The image size is 3750 x 1580 pixels. And the land cover types labeled here are 19 types, mainly cash crops. （3）Classification result：With the first three principal components of the spectrum and its corresponding eight spatial texture features and the vegetation index, such as NDWI and NDVI, the aerial hyperspectral remote sensing image of Matiwan Village in Xiongan New Area were classified using random forest classification algorithm. The total classification accuracy is 97%, and the kappa coefficient is 0.98. According to the confusion matrix, the confusion of Robinia pseudoacacia, Pear tree and Acer complex is serious. that cause the classification accuracy of Robinia pseudoacacia is low. （4）Conclusion：An aerial hyperspectral remote sensing dataset of Xiongan New Area (Matiwan Village) with high spatial resolution and spectral resolution was shared in this study. The dataset was classified using random forest classification algorithm, and the total classification accuracy is 97%. It shows that the dataset can provide good data support for hyperspectral classification research, and also can serve for the design and demonstration of hyperspectral imaging spectrometer.