4D radar is renowned for its all-condition perception and cost efficiency, yet its full potential in target detection remains underexplored due to the sparse nature of its point cloud data. This sparsity often leads to reduced precision in existing 3D detection methods. To address this issue and enhance the feature encoding of 4D radar point clouds, we have developed a novel detection methodology named PA-Net. PA-Net introduces a dual-channel point cloud attention mechanism, PCDC-Attention, which is specifically tailored for pillarized point clouds. This technique enables PA-Net to perform multi-angle pooling on pillarized point cloud data, building upon the PointPillars framework. Utilizing a feedforward neural network with shared weights, the method dynamically generates adaptive attention weights that are applied to point cloud features before pooling. This significantly refines the extraction of quantitative and channel features from the point cloud, thereby markedly improving detection precision. Assessment on the publicly accessible VoD dataset reveals that PA-Net substantially surpasses other point cloud-based object detection methods in accuracy. Notably, it achieves an average enhancement in detection accuracy of 4.2% over the established PointPillars baseline, showcasing the effectiveness of PA-Net in elevating object detection capabilities with 4D radar data.