Camouflaged object detection (COD) aims to accurately segment objects that are visually concealed within their surroundings. Owing to the inherently ambiguous boundaries between camouflaged objects and their backgrounds, as well as their high visual similarity, COD presents significantly greater challenges compared to conventional visual detection or segmentation tasks. Although existing methods have achieved remarkable progress across various scenarios, they often suffer from the loss of fine-grained details, particularly in complex scenes. To address this limitation, we propose a bidirectional semantic fusion architecture for COD. In contrast to existing approaches, our method achieves a more balanced integration of low-level structural features and high-level semantic representations, enabling more precise preservation and enhancement of detailed information. Specifically, the proposed architecture adopts a dual-path fusion strategy: (1) a top-down path, where high-level semantics are progressively transmitted and fused into low-level features to improve detail precision and object localization accuracy; and (2) a bottom-up path, which starts from low-level features and progressively refines high-level representations to ensure that fine-grained local details are effectively preserved. Extensive experiments conducted on three benchmark COD datasets validate the effectiveness of the proposed method. Our model consistently outperforms state-of-the-art approaches across four widely adopted evaluation metrics, demonstrating its robustness, accuracy, and superiority in camouflaged object detection.