<p>In resource-constrained environments like embedded devices, unmanned platforms, and edge computing systems, lightweight camouflage object detection (LCOD) is critical for efficient and accurate target detection, as it effectively facilitates the extraction of discriminative features in challenging scenes where the target is visually blended into the background. Existing LCOD models reduce computational demands but often struggle to balance detection accuracy and parameter efficiency in complex scenarios. To address this, we propose ULCOD-Net, an ultra-lightweight COD framework integrating gate-based multi-feature fusion and dual-constraint (including boundary and region). Specifically, we introduce a lightweight boundary-region decoder (LBRD) to leverage initial region and boundary cues, enhancing object localization. A gate-based multi-level feature fusion module (GMFFM) enables multi-level feature interaction via an attention-based gating mechanism, improving global information propagation and compensating for the limited capacity of lightweight networks. Additionally, a region-constrained feature refinement module (RFRM) progressively refines multi-layer features to produce high-quality camouflage maps. Extensive experiments on four benchmark datasets demonstrate that ULCOD-Net, with only 2.5 million (M) parameters and 3.1 giga (G) computational complexity, achieves F-measure scores of 0.837, 0.758, 0.714, and 0.787 on CHAMELEON, CAMO, COD10K, and NC4K, respectively, outperforming existing lightweight COD models and even surpassing several state-of-the-art heavyweight methods. These results highlight ULCOD-Net’s significant potential for real-time application in resource-limited settings.</p>