<p>The precise and efficient segmentation of polyps is essential for improving the accuracy and effectiveness of colonoscopy screenings. Currently, Transformer-based models, are extensively utilized in the area of polyp segmentation and have produced outstanding results due to their strong ability to capture long-range dependencies. However, capturing long-range dependencies can, to a certain degree, divert the model’s attention away from local features, affecting model performance. Additionally, some existing methods have employed overly heavy decoders to improve segmentation accuracy, leading to unnecessary resource usage. To tackle these challenges, we propose a Receptive Aggregation Network that includes a Receptive Aggregation Module (RAM) and a Hierarchical Semantic Decoder (HSD). The RAM adaptively selects and aggregates local and global attention, reducing information redundancy and providing the necessary receptive fields for multi-scale feature fusion, thereby improving segmentation performance. The HSD progressively fuses abstract features from the deepest layers with shallow features in a top-down manner, achieving global reasoning across all scales and enabling the model to achieve good performance with fewer parameters. Extensive experiments on five widely-used benchmark datasets demonstrate that our method can robustly segments polyp images of various shapes and sizes under different conditions, achieving competitive segmentation performance with relatively low parameter sizes and computational complexity compared to current mainstream methods.</p>