To address the challenges in the impact resistance optimization of lattice-filled structures—specifically the high dependence on subjective initial partitioning when reducing design variables in parametric optimization, and the insufficient precision of discrete design variables in Hybrid Cellular Automata (HCA)-based topology optimization—this study proposes a topology-parameter collaborative optimization method. Adopting a "topology-first, parameter-second" strategy, the method aims to minimize the maximum displacement response of the structure under impact loading. Initially, topology optimization based on the HCA method is utilized to rapidly explore the optimal spatial distribution of materials within the discrete design space, using element strain energy homogenization as the criterion. Subsequently, the material distribution resulting from topology optimization serves as a "smart partitioning" template to replace traditional K-means clustering based on the initial design. This approach achieves effective dimensionality reduction of high-dimensional design variables, endowing the partitions with clear physical significance while overcoming the limitations of partitioning strategies in parametric optimization. Finally, based on the established partitions, a refined parameter optimization for the strut diameters within each zone is conducted using continuous variables, integrating a Radial Basis Function (RBF) surrogate model with the FLT-Net fast prediction method. Validated through typical case studies including sandwich plates, cylinders, and flanged ring structures, the results demonstrate that this collaborative optimization method effectively combines the configuration exploration capability of topology optimization with the numerical fine-tuning advantages of parameter optimization. Compared to the initial design and single-method optimization, the proposed approach significantly reduces the maximum displacement response, yielding gradient lattice-filled structures with superior impact resistance.