The low signal-to-noise ratio and fuzzy morphology of infrared (IR) small targets pose certain challenges in the research of such target segmentation tasks in complex backgrounds. To better separate small targets from clutter backgrounds, an innovative multi-scale cascaded fusion network (MSCFNet) is proposed. Specifically, MSCFNet preserves and utilizes small target information to the maximum extent through the multiple interaction between multi-scale features. At the same time, a feature enhancement module is designed to effectively extract and integrate target information from global semantic and local context, improving the discriminability of targets and complex backgrounds. The experimental results prove that MSCFNet can effectively segment IR small targets in various complex environments and exhibits better performance on two publicly available IR small target segmentation datasets.