EFNet: Multiscale Edge Fusion Network for Camouflaged Object Detection

Camouflaged object detection (COD) aims to segment objects that are seamlessly integrated into their surroundings. Existing COD methods face challenges in processing multi-scale and multi-level information, and they often lack the capability to effectively fuse fine-grained details. To address these issues, we propose a Multiscale Edge Fusion Network (EFNet). Specifically, we design a simam-based edge prior (SEP) module to extract edge features while preserving object structures. Subsequently, we present a multiscale edge-enhanced (MEE) module that refines and integrates multiscale edge information, thereby enhancing the model’s sensitivity to object boundaries. Furthermore, we propose a multiscale semantic detail fusion (MSDF) module, it injects semantic features into low-level representations and refines high-level features with finer details. Leveraging the effectiveness of these modules, EFNet demonstrates an enhanced ability to accurately identify camouflaged objects. Experiments conducted on multiple datasets validate the efficiency and superiority of the proposed EFNet.