The detection of infrared small target plays a vital role in defense and surveillance scenarios. However, the small size of the target coupled with the complex and dynamic background presents substantial challenges in detection tasks. Traditional methods, reliant on hand-crafted features, suffer from poor adaptability and generalization, while deep learning approaches struggle with extreme target-background imbalance and limited target features. To overcome these limitations, we propose the Dual-Stream Background-Target Decoupling Network (DBTNet), a groundbreaking architecture uniquely crafted for efficient detection of small targets in infrared imagery. DBTNet employs a dual-stream design: a background stream with a Global Enhancement Module (GEM) captures large-scale contextual cues for effective background suppression, and a target stream with a Local Enhancement Module (LEM) enhances subtle target features. These streams are integrated via an Adaptive Feature Fusion (AFF) mechanism, which dynamically balances information across scales using spatial and channel attention. Our contributions include the dual-stream framework, GEM and LEM for computational efficiency, and the AFF module for optimized feature integration. Comprehensive evaluations on benchmark datasets reveal that DBTNet attains cutting-edge performance, significantly improving detection accuracy and reducing false alarm rates. This approach resolves the intrinsic conflict between suppressing the background and enhancing the target, significantly boosting detection performance in challenging situations.