AIFENet: Attention-Induced Feature Enhancement Network for Infrared Small Target Detection

The detection performance of infrared small targets is an important factor restricting the development of infrared search and tracking systems. However, existing infrared small target detection algorithms have weak processing capabilities for target shape and structure information, resulting in inaccurate target segmentation results. To address the above challenges, this paper proposes the attention-induced feature enhancement network (AIFENet) model. The model is mainly composed of the bifurcation convolution enhancement (BCE) module and the attention-induced feature adaptation (AIFA) module. The BCE module combines the bifurcated convolution block (BCB) module and the basic channel attention (BCA) module, which work together in the extraction process of shallow features to ensure that the shallow features fully contain the information required for positioning and segmentation, and realize feature information of precise optimization. The AIFA module achieves information interaction between deep features and shallow features through the spatial feature reconstruction (SFR) module and the enhanced channel attention (ECA) module, thereby obtaining more refined semantic information. Through extensive experimental verification, the AIFENet model shows significant advantages in dealing with the above problems, and is significantly better than the baseline method in detection performance.