Lite-PolypNet: A Lightweight and Efficient Network for Polyp Segmentation in Colonoscopy Images

Early detection of colorectal polyps is of great significance in preventing colorectal cancer. However, existing segmentation methods often struggle to balance accuracy and computational efficiency. To address this issue, this paper proposes a lightweight and efficient polyp segmentation network named Lite-PolypNet. Built upon MobileNetV3 as the backbone, the network integrates a progressive feature aggregation module, a global attention augmentation module, and a dual-branch decoder structure to effectively fuse multi-scale features and global contextual information, thereby enhancing boundary reconstruction and small polyp detection capabilities. Extensive experiments conducted on five public datasets demonstrate that Lite-PolypNet achieves high segmentation accuracy (with a maximum Dice score of 94.7%) while significantly reducing model parameters and computational complexity. Compared with representative baseline models, Lite-PolypNet reduces the number of parameters by a factor of more than six and significantly decreases the FLOPs, making it suitable for deployment in resource-constrained environments.