MSEE: Multi-scale Edge Enhanced Algorithm for Infrared Dim-Small Target Detection in Complex Background

Capitalizing on the inherent anti-interference capabilities of infrared technology, infrared target detection finds extensive application in precision guidance systems, border surveillance, and disaster early-warning systems. However, targets in these scenarios usually exhibit low signal-to-noise ratios, small sizes (typically <5 × 5 pixels), and susceptibility to background clutter, presenting substantial obstacles for the detection of infrared dim-small target (IRDST). This paper augments YOLOv11 with a P2 level feature map high-resolution detection head to enhance small target detection. We propose the C3K2-MSEE (Multi-Scale Edge Enhanced) module for precise infrared target detection, which enhances target contours and fuses multi-scale features by extracting local information. By integrating Shape-IoU (Intersection over Union) into the Normalized Gaussian Wasserstein Distance (NWD), we improve bounding box similarity measurement. The optimized model yields 15.3% and 2% gains in mAP@50 and mAP@50–95 respectively over baseline YOLOv11, demonstrating efficacy in extracting discriminative features while suppressing background interference for IRDST detection.