A unified framework for small object analysis in UAV imagery: Tracking single target from multiple detected targets
image: Schematic diagram of the CTDT framework for detection and tracking: When there is no prior location, the detection stage uniformly crops the original remote sensing image into small-size images. Small-size images are randomly sampled without repetition until all targets are detected or all small-size images have been sampled. For tracking, a primary search region is selected based on the prior positions. If the target is not found in the primary search region, the search area is gradually expanded until the target is found or the entire image is covered.
Credit: Chinese Journal of Aeronautics
Detecting and tracking small targets in UAV remote sensing images is a key task in border patrol, traffic monitoring, disaster rescue, and other fields. In typical UAV remote sensing scenarios, targets like vehicles and ships occupy very few pixels in the image. At the same time, the relative movement between the UAV and the target may cause the target to move quickly, be obscured, or even disappear briefly in images, posing challenges for real-time, high-precision detection and tracking. Existing deep learning methods still face the following issues when directly applied to small targets in remote sensing images: downsampling high-resolution remote sensing images tends to lose detailed information of small targets; Network parameters are too large, making it difficult to run in real time on onboard computers; Poor adaptability to rapid target movement, temporary disappearance, and image quality degradation.
Prof. Yanfang Liu from Harbin Institute of Technology recently published an article in the Chinese Journal of Aeronautics (https://doi.org/10.1016/j.cja.2025.103917) proposing a unified framework for real-time detection and tracking of small UAVs, called Cropped Target Detection and Tracking (CTDT), which is expected to solve the above problems.
The CTDT framework is based on cropping and random sampling strategy: high-resolution images are evenly cropped into multiple small-sized images, from which a random batch is constantly selected without repetition and fed into a network to locate the target until the target is discovered or all small-sized images are used. The strategy avoids traversing all small-sized images.
To enhance adaptability to fast movement targets, the authors designed multi-level search areas. If no target is found in the current area, the circle expands step by step to the next search area and continues searching until the target is located or the entire image is covered. This ensures real-time performance while effectively handling situations where targets move out of their initial range or reappear after brief disappearance.
To accurately and efficiently detect incomplete targets from small-sized images, the authors design a CTFE network and explore the effects of factors such as target integrity, target pixel count, and cropped image size on CTFE network accuracy, providing theoretical support for network optimization and practical deployment.
Their experimental results showed that CTDT achieved high efficiency while ensuring tracking accuracy, while also showing stronger robustness against rapid target movement and transient loss. Professor Yanfang Liu stated that in the future, by optimizing the CTFE network structure, the method will be expanded to dense multi-object tracking and integrated multimodal observations to improve performance under low light and extreme weather conditions.
Original Source
Yanfang Liu, Jiayu She, Yituo Song, et al, Naiming Qi. A unified framework for small object analysis in UAV imagery: Tracking single target from multiple detected targets [J]. Chinese Journal of Aeronautics, 2026, 39(5):103917, https://doi.org/10.1016/j.cja.2025.103917.
About Chinese Journal of Aeronautics
Chinese Journal of Aeronautics (CJA) is an open access, peer-reviewed international journal covering all aspects of aerospace engineering, monthly published by Elsevier. The Journal reports the scientific and technological achievements and frontiers in aeronautic engineering and astronautic engineering, in both theory and practice. CJA is indexed in SCI (IF = 6.9, Q1), EI, IAA, AJ, CSA, Scopus.