<p>Population spatialization aims to derive the spatial distribution of population at fine scales using census data, serving as a critical underpinning for sociology, geography, and urban–rural planning. Current studies often rely on a single model to generate the fine-grained spatial distribution of population. However, owing to the evident disparities in regional characteristics and geographic data between urban and rural areas, a unified or global model fails to accurately reveal population distributions across heterogeneous regions. Consequently, this study proposes a divide-and-conquer method for predicting the fine-grained spatial distribution of population: urban populations are predicted using a two-level extra trees model, while rural populations are estimated via deep learning-based building area extraction and spatialization through building area ratios. The experimental results obtained using the proposed method in Xiangtan and Changsha, China, indicate that the coefficients of determination (<i>R</i>²) are 0.889 and 0.936, respectively, and the root mean square error is 11,852 and 9636, respectively, in the urban area of Xiangtan and Changsha, outperforming comparative methods. Similarly, the <i>R</i>² and RMSE are 0.806 and 9036, respectively, in the rural area of Xiangtan, and 0.835 and 10,040, respectively, in the rural area of Changsha. The statistical results of the overall accuracy evaluation validate the effectiveness of the proposed method.</p>