<p>Camouflaged object detection (COD) aims to accurately segment objects visually concealed within complex backgrounds, which is a fundamental yet challenging task in computer vision. However, existing frequency-based methods suffer from two major limitations: first, they often rely on a single transform (e.g., Fourier or wavelet alone), making it difficult to simultaneously capture global structures and local details; second, even when multiple frequency representations are involved, they lack effective cross-domain fusion with spatial information, leading to suboptimal boundary localization and detail recovery. To address these challenges, this paper presents a Dual-Frequency–Spatial Collaborative Network (DFCNet) for COD, which systematically integrates and adapts complementary frequency-domain representations within a spatial feature extraction framework. Specifically, it comprises two key components: (1) an attention-guided Fourier-wavelet enhancement module that effectively combines global structural representations from Fourier transform with local spatial-frequency correspondences from wavelet transform—adapting their complementary strengths to enable full-scale perception from macro contours to micro textures; (2) a Visual Group Mamba (VGM) decoder that efficiently integrates frequency-enhanced features with spatial representations via multi-directional scanning, substantially improving boundary localization and detail recovery. Through the collaborative learning of these modules, DFCNet generates prediction results with complete structures and sharp boundaries. Experimental results on CAMO, COD10K, and NC4K datasets demonstrate that DFCNet consistently outperforms 21 state-of-the-art methods across key metrics.</p>