In order to speedup the reconstruction of particle hologram, a fast reconstruction method based on two-level parallel framework was proposed. The framework combines open multi-processing(OpenMP) and compute unified device architecture(CUDA). The image level parallelism was realized by OpenMP, and the pixel level was realized by CUDA. The parallel computation was carried out on each reconstruction plane and pixel. Taking the coal particle hologram as the test object, the single thread reconstruction program and the two-level parallel reconstruction program were used to reconstruct the hologram at the same time. The reconstruction results and time consumption of the two methods were compared. The experimental results show that the results of two-level parallel reconstruction are consistent with those of single thread reconstruction, and the time consumption can be greatly reduced. For a high-resolution 5000×5000 hologram, the speedup ratio is up to 48.3 with 40 reconstruted planes.This computing framework has a good application prospect in the real-time and on-line holographic diagnosis of particle field.