Geostationary satellites observe the earth surface and atmosphere with a short repeat time which can thus provide aerosol parameters with high temporal resolution. Due to the limited information content in satellite data, and the coupling between the signals received from the surface and the atmosphere, the accurate retrieval of multiple aerosol parameters over land is difficult. Here we propose a Neural Network AEROsol retrieval framework for Geostationary satellite (NNAeroG) which can potentially be applied to different instruments to retrieve various aerosol parameters. NNAeroG was applied for aerosol retrieval using data from the Advanced Himawari Imager on Himawari-8 and the results were evaluated versus independent ground-based sun photometer reference data. The retrieved Aerosol Optical Depth, Ångström Exponent and Fine Mode Fraction are significantly better than the official JAXA aerosol products. The use of thermal infrared bands is meaningful for aerosol retrieval.