Aerosol optical thickness (AOD) is an important characterization parameter of aerosol concentration and atmospheric turbidity. Inversion of atmospheric AOD by remote sensing is an important way in the process of atmospheric monitoring and management, and in which the selection of methods suitable for the imaging characteristics of satellite sensors and the type of aerosols in line with the study area has always been the focus and difficulty of AOD inversion. In view of the problem that the traditional dark target method can not be directly applied to the multispectral remote sensing data of Gaofen Ⅳ (GF-4) satellite, this paper studies the distribution of the red and blue band equivalent surface reflectivity in GF-4 multispectral data and the linear relationship between them, and improves the dark target method to make it suitable for GF-4 satellite multispectral remote sensing data in combination with AOD inversion principle. The effect of input parameters on AOD inversion accuracy in the 6S radiation transfer model was analyzed, and the experimental results showed that aerosol type is one of the key factors affecting the high-precision inversion of AOD. The samples of aerosol characteristics in Beijing-Tianjin-Hebei area was analyzed by particle swarm optimization (PSO) cluster algorithm, by analyzing the proportion and half-life changes of the clustering results of each aerosol type, the C1 and C4 aerosol types in cluster results and the continental aerosol type of 6S models are finally determined to invert the AOD in Beijing-Tianjin-Hebei region. The inversion results were compared with MODIS aerosol products and AErosol RObotic NETwork (AERONET) ground-based site data, and the suitability and characteristics of different aerosol types are evaluated by evaluation criteria such as correlation coefficient and absolute error. The experimental results show that the C4 aerosol type, which is dominated by fine particles, is more satisfied with the characteristics of aerosols in the summer and autumn of Beijing-Tianjin-Hebei, and has better consistency with AERONET ground-based data. It is further proved that the PSO clustering algorithm can effectively reduce the influence of aerosol type difference on AOD inversion accuracy.