Research on Infrared Dim and Small Target Detection Based on U-Net

To address the challenges of low accuracy and high false alarm rate in infrared small target detection, this paper proposes an improved method based on the U-Net architecture. The proposed approach leverages cascaded channel and spatial attention modules to capture subtle infrared target features, enabling progressive feature interaction and adaptive feature enhancement. These attention modules are integrated into the feature pyramid fusion module, which efficiently extracts feature information across multiple scales. By iteratively fusing and enhancing these features, the method effectively incorporates the contextual information of small targets, thus improving the model's ability to process and extract deep features. Experimental results demonstrate that, compared with the Transformer-based detection method, the proposed method improves Intersection over Union (IOU) and Probability of Detection (POD) by 1.16% and 1.40%, respectively, while reducing the False Alarm Rate (FAR) by 19.16%. These results validate the effectiveness and practicality of the proposed method, highlighting its broad application potential in infrared small target detection. The method's performance improvements indicate its suitability for real-world scenarios and suggest that it could be an effective tool for various applications that require precise detection in complex environments.