Infrared Small Target Detection Algorithm Based on Super-Resolution Reconstruction and Improved YOLOv8n

Infrared small target detection primarily involves the recognition and localization of small targets within infrared images, holding great potential for various applications. This task faces several technical challenges due to the small size of targets in infrared images and the complexity of their backgrounds, including the low signal-to-noise ratio and limited pixels of the targets. We propose an improved YOLOv8n detection model based on sample super-resolution reconstruction to address these issues. First, the Spatially-Adaptive Feature Modulation (SAFMN) super-resolution method is employed to improve the resolution of the sample images and enhance the texture features of the targets. Subsequently, the processed images are cropped into smaller patches, and only patches containing targets are retained to strengthen the features of the infrared small targets and reduce the interference of the background. Finally, we integrate a parallelized patch-aware attention (PPA) module into the YOLOv8n network, which uses a multi-branch feature extraction strategy to capture features at different scales and levels of the targets. Experimental results show that, compared to the state-of-the-art detection methods, our method improves the mAP50 by 2.4% and Precision by 6.3% on the IRSTD1k dataset.