Infrared small target images are usually interfered by thermal and speckle noise, and their detection accuracy and robustness face severe challenges. Under high-noise conditions, issues such as low signal-to-noise ratio, background clutter, and blurred target structures severely hinder detection accuracy.In this paper, we propose IR-SDTNet, a denoising detection framework for infrared small targets. It adopts a hierarchical framework and enhances feature representation through dual modeling of local and global features. A multiscale feature fusion mechanism combined with a jump connection integrates the encoder’s multi-level features with the decoder’s up-sampling features, which enables noise reduction and detail recovery. Experimental results show that IR-SDTNet not only eliminates most noise from infrared images but also helps restore fine details, and furthermore enhances the detectability of small targets, especially under high-noise backgrounds where its denoising and restoration capabilities are particularly strong.