HSOD-BIT is the first large-scale, high-quality hyperspectral salient object detection benchmark dataset, which aims to leverage the advantages of spectral information to achieve higher accuracy in salient object detection tasks. Targeting the data needs of contemporary deep learning models, this dataset provides pixel-level manual annotations for 319 hyperspectral data cubes and generates corresponding pseudo-color images. Each data cube contains 200 bands, covering spectral information from visible light to near-infrared bands, with a spatial resolution of up to 1240×1680 pixels. In addition to conventional scenes, the dataset also specifically collects challenging data to reflect the complexity of the real world, such as similar background interference, uneven lighting, overexposure and other challenging scenes. This further enhances the practicality and evaluation capabilities of the dataset.