Infrared images acquired under complex background conditions are often affected by background clutter, local high-response interference, and non-uniform fluctuations, which may reduce target saliency and local discriminability. To address this issue, this paper proposes an improved UNetFPN-based feature enhancement network, termed CBAM-UNetFPN. Built on an encoder-decoder architecture, the proposed method introduces a feature pyramid fusion mechanism to combine shallow spatial details with deep semantic information, and incorporates an attention enhancement strategy to enhance target-related responses while suppressing redundant background activations. Experiments were conducted on three public infrared image datasets, namely NUDT-SIRST, IRSTD-1k, and WideIRSTD-Weak, and the enhancement performance was evaluated using the signal-to-clutter ratio, background suppression factor, and contrast gain. The results show that the proposed method achieves stable enhancement performance across scenes with different levels of complexity. Comparative experiments further indicate that CBAM-UNetFPN can better balance target response enhancement and background suppression under complex background conditions, thereby improving the local discriminability between target regions and the surrounding background.