Although Transformers have made significant progress in 3D point cloud processing, efficiently and accurately learning valuable low-frequency and high-frequency information to improve applications remains a challenge. This paper proposes a novel point cloud learning network, termed Multi-Scale Dual-Branch Dual-Attention Network. First, during the extraction of point clouds, compared to methods that search for neighboring points at fixed scales, the use of a multi-scale KNN approach not only preserves local structural details but also more effectively captures global geometric information. Secondly, existing methods mostly focus on local spatial information while neglecting global spatial information, leading to information loss. This paper introduces a dual-branch dual-attention architecture to extract different spatial features, proposing a Local Window Attention and Global Channel Affinity Attention mechanism to respectively extract low-frequency and high-frequency information. On this basis, the GR-KAN layer is introduced in the classification head, replacing the traditionally used MLP layer, enabling more flexible handling of nonlinear features, thus making the network more sensitive to complex dataset. Extensive experiments demonstrate that the proposed model achieves 93.8% accuracy on the ModelNet40 dataset and 86.5% accuracy on the ScanObjectNN dataset, showcasing its superior performance and broad application prospects in 3D point cloud processing.