Particle-based simulation has a wide range of applications across various manufacturing processes. As one of the key components of particle-based methods, generating initial particles to accurately characterize all the geometry features is highly challenging. Unlike the currently in-trend particle generation methods based on implicit geometries by constructing a level-set, in this work, we present a feature-preserving particle generation method for arbitrarily complex objects directly from explicit geometries. The main advantage of the proposed method is twofold. First, the generated particle profile recovers the input geometry exactly, including singularities and sharp features. Second, the algorithm is more lightweight in terms of memory footprint, since the construction of a global signed distance function is not required. We have applied our method for generating particles on both geometry surfaces and volumes enclosed by a set of user-defined CAD models. To achieve the above goals, a set of algorithms are presented. First, a points projection algorithm is proposed to characterize the features from input geometries, along with a feature edge capturing algorithm. Second, a parallel memory mapping algorithm corresponding with a sparse-space encoding data structure is developed to reduce memory consumption and increase efficiency. Lastly, a new mechanism is introduced to update particle positions based on explicit geometry definitions, which allows for efficient optimization of undesired particle distribution both in the volume and on the surfaces. A set of benchmarks, ranging from dragon, gearbox, automobile and airplane, are tested to demonstrate the performance of the present method. The results exhibit that the performance and the accuracy of describing the geometry are significantly improved. Numerical tests including industrial-level benchmarks are carried out in the end to validate the accuracy and performance of the proposed particle generation method. The results verified that the present method is effective and robust for particle-based simulations.