Abstract: Infrared dim-small target detection technology has significant practical value and importance in applications such as air defense, target guidance, and aerospace. This study addresses key challenges, including the limited information content of infrared dim small targets, complex background interference, and small target sizes, by proposing a multi-scale feature enhancement-based detection method for infrared dim small target detection. Using the YOLOv8 network model as a baseline, a U-shaped structure that integrates pooling operations with dilated convolutions is designed within the backbone to fully retain and extract small-target features while effectively suppressing surrounding noise. In addition, an enhanced neck module, namely, the Information Gathering and Reparameterization module, is introduced to enable sufficient fusion and enhancement of multi-level and deep features extracted by the backbone. To enhance attention to targets of varying scales and improve their discrimination from complex backgrounds in infrared data, a channel-prior convolutional attention mechanism is introduced and integrated into the C2f structure. Experimental results on the augmented NUAA-SIRST dataset show that, compared with the YOLOv8n model, the proposed model achieves improvements of 13.7%, 8.7%, and 5.6% in precision(P), average precision (mAP₅₀), and mAP_50-95, respectively. Furthermore, in comparison with other state-of-the-art methods, the proposed model attains superior detection performance, reaching an optimal level.