Infrared Dim Small Target Detection Based on Regional Refinement Network

Target detection of infrared images has been extensively utilized in various fields, including early warning, guidance, and monitoring. However, the performance of infrared target detection is hindered by target characteristics of dim and small. For infrared dim small objects, there are issues of a small proportion of pixels, serious background interference and lack of visual information. To solve these problems, an infrared dim small target detection model based on the regional refinement network is proposed in this paper. In the proposed framework, a feature extraction module based on the feature pyramid with multi-scale receptive fields is proposed to extract spatial information from infrared images firstly. Then, a regional feature refinement module is proposed to improve the difference between target and background. Experiments are conducted on the SIRST dataset to prove the effectiveness of the proposed network. It can be seen the method we proposed is able to yield superior results on infrared dim small target detection.