The physics of atomization process involve many spatial scales, generating a wide variety of liquid inclusions of different sizes with large density and viscosity ratios between liquid and gas phases. To correctly capture the dynamics of these phenomena, each scale should be resolved with an appropriate method to ensure the conservation of physical quantities (mass, momentum) as well as the jump conditions across the liquid-gas interface. To address these problems, an original multi-scale methodology has been developed. It consists of a core coupled Level-Set/Volume-of-Fluid method (CLSVOF) for accurate capture of primary atomization, an adaptive mesh refinement technique (oct-tree AMR) to dynamically optimize the structured Cartesian mesh and a particle tracking algorithm to capture droplet dynamics. An improved Eulerian-Lagrangian coupling has been developed to assure a smooth transition between the Eulerian and the Lagrangian modelling of the droplets, where both methods approach their design limits. The overall procedure is tested on simplified numerical tests and validated on a realistic planar liquid sheet atomization case. Results show its ability to reproduce the whole atomization process, from large scale instabilities to small droplet dynamics, and allow a preliminary statistical spray analysis.