This paper presents a compressive hyperspectral video imaging system using a single-pixel detector, achieving high-throughput hyperspectral video recording at a low bandwidth. The system leverages the compressibility of 4D hyperspectral videos, encoding the scene into highly compressed measurements while maintaining temporal correlation. A reconstruction method based on signal sparsity in 4D space and deep learning is proposed, significantly reducing reconstruction time. Experiments demonstrate the system's ability to reconstruct 128 × 128 hyperspectral images with 64 spectral bands at over 4 frames per second, offering a 900× data throughput compared to conventional imaging. The system's advantages include the elimination of the need for 2D array sensors, enabling imaging in non-visible wavelengths, and reducing storage and transmission requirements, making it suitable for resource-constrained applications such as satellites and rovers.This paper presents a compressive hyperspectral video imaging system using a single-pixel detector, achieving high-throughput hyperspectral video recording at a low bandwidth. The system leverages the compressibility of 4D hyperspectral videos, encoding the scene into highly compressed measurements while maintaining temporal correlation. A reconstruction method based on signal sparsity in 4D space and deep learning is proposed, significantly reducing reconstruction time. Experiments demonstrate the system's ability to reconstruct 128 × 128 hyperspectral images with 64 spectral bands at over 4 frames per second, offering a 900× data throughput compared to conventional imaging. The system's advantages include the elimination of the need for 2D array sensors, enabling imaging in non-visible wavelengths, and reducing storage and transmission requirements, making it suitable for resource-constrained applications such as satellites and rovers.