Gas-solid two-phase flow is widely encountered in energy and chemical industries and the interphase drag force is believed to be the dominant factor affecting the flow. Although the homogenous microscopic drag models, e.g. the BVK drag law (AIChE Journal, 2007, 53(2): 489-501), could accurately predict the drag force for homogenous regions, i.e. the dilute or dense phases, they failed to predict the drag at the interface of dense and dilute phases where the surface of particle clusters was located. To study the influence of the interface on the drag force, the particle-resolved direct numerical simulation (PR-DNS) was performed on the process of fluid flowing through different particle clusters at low Reynolds numbers. The results showed that at the interface of dense and dilute phases, the predictions of the BVK drag model were significantly different from the PR-DNS results. Meanwhile Chen et al.'s model (Int. J. Multiphase Flow, 2020, 128:103266), which considered the influence of the interface, cannot accurately predict the drag force where the solid holdup of the dilute phase was not zero. Therefore, this work proposed a method of decomposing the mesh near the interface to predict the drag force. To validate the proposed model, PR-DNSs of flow past various particle clusters, were performed and the predictions of several drag force models were calculated. It was found that the new model not only had a similar predictability with Chen et al.'s model when the dilute phase solid holdup approaches 0, but also had better Pearson correlation coefficients and fitness than the Chen et al.'s model when the solid holdup of the dilute phase was not zero. In conclusion, the proposed drag model could accurately account for the influence of the interface on drag force with different dilute and dense solid volume fractions, and it recovered the BVK law when the local gradient of volume fraction approached zero.