The invention discloses a quick reconstruction method of a double-camera spectral imaging system based on a GPU, and relates to a method which can quickly acquire a high-resolution hyperspectral image, wherein the method relates to the field of computational photography. The method is applied on a double-camera spectral imaging system based on coded aperture snapshot spectral imaging and a gray-scale camera. A hyperspectral image reconstruction problem is converted to a plurality of sub optimization problems, and furthermore a GPU is utilized for finishing solving of each sub problem. A cuBLAS database and a conjugate gradient reduction method are utilized for updating the hyperspectral image. A soft-threshold function is utilized for updating an auxiliary variable. Iteration is performed for finishing reconstruction of the hyperspectral image. The method of the invention can realize high-quality hyperspectral image reconstruction of the double-camera spectral imaging system and furthermore has advantages of ensuring high spatial resolution and high spectral fidelity of a reconstruction result, greatly improving reconstruction efficiency of the hyperspectral image, and expanding application range of the hyperspectral image. The quick reconstruction method can be used in a plurality of fields of manned space flight, geological exploration, vegetation studying, etc.