The finite element method (FEM) is the most widely adopted numerical framework for solving mechanical equilibrium problems in polycrystalline representative volume elements (RVEs). It is commonly employed to investigate the influence of microstructural features on micromechanical fields and to compute effective properties via numerical homogenisation, thereby establishing structure-property relationships. In polycrystalline materials, micromechanical fields are highly heterogeneous, with pronounced localisation near grain boundaries, necessitating locally refined discretisation. Although nonuniform and unstructured FEM meshes can partially address this requirement, it remains challenging to generate periodic unstructured meshes for complex polycrystalline microstructures. Furthermore, the accurate representation of internal boundaries or evolving interfaces typically demands fine meshes and complex remeshing strategies. Octree (or quadtree in two dimensions) meshes provide an efficient alternative by enabling rapid mesh transitions and adaptive refinement, with fine elements near grain boundaries and coarser discretisation in grain interiors, thereby reducing the overall degrees of freedom. However, the presence of hanging nodes prevents their straightforward use within conventional FEM. The scaled boundary finite element method (SBFEM), which naturally supports polygonal and polyhedral elements, overcomes this limitation and enables the direct use of octree-based discretisations. In this work, an octree-based SBFEM framework is employed to simulate elastoplastic deformation in polycrystalline RVEs using crystal plasticity. The proposed approach is systematically compared with conventional FEM in terms of micromechanical field distributions and computational efficiency. The results demonstrate that octree-based SBFEM accurately captures the heterogeneous local fields characteristic of polycrystalline materials while significantly reducing the computational cost.