With the increasing application of deep learning in medical imaging, polyp detection in intestinal endoscopy has emerged as a critical task to improve diagnostic accuracy and efficiency. However, differences in polyp size and shape make feature extraction difficult. Furthermore, the presence of complex textures in the intestinal wall may cause polyps to blend into the background or be obscured, further complicating feature extraction. To address these issues, we propose a lightweight and efficient detection model LPI-YOLO based on the YOLOv8n model. Firstly, we design a Multi-scale Perceptual Attention (MPA) module that integrates multi-head attention mechanisms with multi-scale perceptual fields, accommodating the diversity of polyp size and shape. Additionally, the Context-Aware Feature Aggregation (CAFA) module is deployed, which incorporates multi-level fine-grained contextual information with refined global channel vectors. This module not only significantly improves the extraction capabilities of polyp features, but also effectively optimizes the complexity of the model. Finally, the point-based up-sampling method DySample is used to replace the original UpSample. The mechanism enhances feature fusion and mitigates the loss of fine-grained details in polyp images. Experimental results indicate that the proposed LPI-YOLO model substantially surpasses the baseline model on the colonoscopic polyp detection dataset. Specifically, it achieves a 6.1% improvement in mAP50, reaching 84.9%, along with enhancements of 8.5% and 6.3% in Precision and Recall, respectively. Moreover, the parameter count is reduced by 18.5%. These results outperform state-of-the-art detection algorithms, thus validating the efficiency of LPI-YOLO in intestinal endoscopic polyp detection.