<p>Camouflaged object detection (COD) aims to identify targets that are highly similar to the environment from complex backgrounds through deep feature extraction and reconstruction. However, existing methods in COD tasks still struggle to simultaneously balance local detail perception and global context modeling, and face challenges of insufficient fine-grained feature representation in the single RGB domain. Recently, state space models, represented by Mamba, have received extensive attention due to their excellent long-range modeling capabilities and lightweight parallel scanning algorithms. To this end, this paper presents MambaCOD, a network for COD based on an improved state space model. Specifically, we design a spatial- and frequency-based Dual-Domain-Enhanced State Space (DSS) module. By utilizing features from different domains, this module enhances the discriminative ability to identify subtle differences between camouflaged targets and backgrounds. The local spatial branch captures local semantics from neighboring positions through parallel multi-layer convolutions, working in conjunction with the state space model’s global modeling capability. The frequency-domain branch learns more discriminative features between objects and the background by applying Fourier Transform to the amplitude and phase spectra. In addition, we propose a Multi-scale Detail Enhancement (MDE) module that incorporates multi-level fine-grained information into Mamba, improving the model’s edge detection ability for camouflaged objects. Experiments show that our method achieves competitive performance in both qualitative and quantitative assessments.</p>