Hyperspectral video target tracking based on pixel-wise spectral matching reduction and deep spectral cascading texture features.

•A novel hyperspectral image dimensionality reduction method called Pixel-wise Spectral Matching Reduction (PSMR) is proposed. The proposed method uses two predefined thresholds to distinguish the search area pixel by pixel. The PSMR method not only compress the dimension of HSIs, but also roughly segment the target and background in the search area;•Based on the segmentation results of PSMR, a coarse location mask, which is used to predict the approximate position of the target in the next frame, is generated. In addition, the generated mask suppresses the texture features of the background;•This paper also proposes a novel Spectral Cascading Texture (SCT) feature that is obtained by combining texture features, extracted using the LBP operator, with the Signal-Noise Ratio (SNR) spectral curve. As a new feature that enhances HVT, SCT contains rich spatial and spectral information;•In order to combine the texture, spectral and semantic information, we propose a feature fusion method called group-pixel joint convolution, which can effectively preserve the local characteristics of SCT features.