Subgrid scale (SGS) structures in large eddy simulations (LES) of turbulent particle-laden flows significantly influence particle dynamics, especially those of small inertial particles. In this study, homogenous isotropic turbulence with Taylor's Reynolds number of 102.3 is generated by a direct numerical simulation (DNS) and a wavelet-based coherent vortex extraction method is implemented to extract the coherent SGS structures and then investigate their effects on particle dynamics, including single-particle and particle-pair statistics. Compared to the classical spectral-filtered DNS (FDNS), which cuts off only the high wavenumber components regardless of the turbulence structures in the SGS motions, the wavelet-filtered DNS (WFDNS) can retain more coherent vortex structures in the SGS flow field with the help of the high compression rate characteristics of wavelet transformation. Comparing the results of WFDNS and FDNS at the identical effective grid number, it can be found that the single-particle statistics are mainly controlled by the macro energy-containing structures, and the SGS coherent vortex structures play important roles in the particle-pair dynamics, including the radial distribution function, radial relative velocity, and collision kernel. Therefore, in view of the characteristic of wavelet filtering that preserves the SGS coherent structure, the wavelet-based structural filter should be particularly suitable for LES modeling of particle-laden flow.