Infrared small target detection with super-resolution and YOLO

Infrared remote sensing imaging plays a crucial role in military observation, nighttime security surveillance, forest fire monitoring, and so on. In these applications, detecting dim small targets has always been a challenging problem, especially in complex backgrounds and low-contrast conditions. Existing model-driven methods usually lack robustness in handling noise and small-size targets. Deep learning-based approaches are heavily dependent on data and have limitations in feature processing and fusion, leading to missed detections and false alarms. To address these issues, we propose a small target detection method for infrared images with image super-resolution technology and deep learning. Firstly, we apply super-resolution image preprocessing and multiple data augmentation to the input infrared images. Secondly, we develop a deep-learning network based on YOLO called YOLO-SR, which incorporates a bottleneck transformer block after the spatial pyramid pooling module in the backbone layer to capture long-range dependencies in the infrared images. We design a C3-Neck module in the neck layer to better extract and fuse spatial and channel information. Experimental results show that the proposed method achieves mAP@0.5 scores of 95.2% on the public datasets and effectively addresses the issues of missed detections and false alarms compared to current state-of-the-art data-driven detection methods.