Smoothed particle hydrodynamics(SPH) is widely applied in engineering simulations. For large-scale simulations, single-resolution SPH method requires high computational resources and time. Multi-resolution SPH method is popular since it significantly reduces computational costs while maintaining simulation accuracy. A necessary step for SPH method is particle generation before simulation. It is challenging to generate multi-resolution particles for arbitrary complex geometry while preserving geometry features and details. We present a novel multi-resolution particle generation method based on explicit geometry. Compared to the trending particle generation methods based on implicit geometry like level-set[1], the main advantage of our method is easily handling sharp interface geometries and less memory consumption. In this work, a tag-based background grid system with an adaptive refining algorithm is proposed to flexibly handle predefined refinement zones. A geometry partition algorithm with a compression data structure is established to store the mapping relationship between background grids and geometry while improving parallel efficiency and decreasing memory footprint. A scheme projecting points from multi-resolution grids to geometry surface is developed, along with a feature line extraction algorithm to accurately represent boundary shapes[2]. A physics-driven multi-resolution relaxation mechanism is introduced to achieve high-quality particle distribution. To validate the accuracy and usability of the present approach, geometries including the Stanford bunny, dragon and vehicle are modeled to generate multi-resolution particles. Furthermore, multi-resolution particles produced by our method are applied to industrial SPH simulations such as vehicle wading, vehicle swimming and vehicle tire splashing. It is verified that the present method is robust and applicable for multi-resolution SPH simulations.