Hyperspectral Image Classification Dataset

Four representative real HSI datasets covering complex agricultural and typical urban scenes, with different sizes, frequency bands, and land cover categories, can fully verify the effectiveness of the method. The detailed information of each dataset is as follows:1) PaviaU University: Collected by ROSIS sensors in Pavia, Italy, with 9 categories, image size of 610 × 340 pixels, spatial resolution of 1.3m, and a total of 103 spectral bands covering the visible to near-infrared range. This dataset represents typical urban scenes, with clear land cover categories but local fragmentation.2) Hanchuan: The Headwall Nano Hypersec sensor was used in Hanchuan City, Hubei Province to collect 16 categories of images with a size of 1217 × 303 pixels and a spatial resolution of 0.109m, including 274 bands from 400 to 1000nm. The dataset is mainly composed of complex agricultural areas, with dense distribution of land features and subtle differences in categories, similar spectral characteristics, and high requirements for fine-grained classification ability.3) HongHu: 22 categories were collected using the Headwall Nano Hypersec sensor in Hubei Province, with an image size of 940 × 475 pixels and a spatial resolution of 0.043m, including 270 bands from 400 to 1000nm. The dense crop types and similar spectral characteristics pose significant challenges for distinguishing land boundaries.4) Houston: Collected by ITRES CASI 1500 sensor, with a total of 15 categories, image size of 349 × 1905 pixels, spatial resolution of 2.5m, covering 144 bands from 380 to 1050 nm. Covering land types such as grasslands, buildings, and roads, the adaptability of the model to complex urban scenes can be tested.