A dataset of semi-synthetic infrared dim small moving targets for detection and segmentation

<p indent="0mm">Infrared dim small moving target detection is a research hotspot in the field of target perception, with wide applications in security monitoring, reconnaissance and early warning. Recently, infrared dim small target detection algorithms based on deep learning have been developed rapidly, placing higher demands on the scale and quality of datasets. However, it is difficult to obtain a large amount of real data with fine-grained annotations. Existing datasets either synthesize targets using Gaussian distribution or sample frames to reduce the workload, resulting in unrealistic targets and discontinuous motion trajectories. To address these issues, we proposed a generative adversarial network with spatio-temporal feature fusion to learn the appearance variation of real targets and generate highly simulated infrared targets. Then, by combining diverse scenes and various target motion trajectories from existing datasets, we finally constructed a large-scale semi-synthetic dataset for infrared dim small moving target detection and segmentation under clutter backgrounds. The generated targets range in size from 4×4 to 9×9 pixels with a local signal-to-noise ratio of −5 to <sc>5 dB,</sc> conforming to the definition of dim and small targets. The dataset covers four scenes (building, vegetation, cloud and water), three cases (one, two and three targets), and provides three annotations (center point, bounding box and mask), as well as two evaluation metrics (instance-level and pixel-level). The dataset contains 200 image sequences, 60,000 images, equally divided into a training set and a test set. It can be used in infrared dim small moving target detection, segmentation and tracking research under complex backgrounds.