Comparison of small object detection approaches in unmanned aerial vehicle (UAV) images
Bibliographic record
Abstract
Accurate detection of small objects remains a challenge in computer vision, especially in the context of unmanned aerial vehicle (UAV) imagery, where objects often appear at low resolutions in complex backgrounds. In this paper, we review the limitations of current deep learning-based object detection algorithms and analyze the performance of recent architectures including YOLO, Mask Region-based Convolutional Neural Network (MRCNN) and adaptations for small object detection in aerial scenarios. We highlight key architectural strategies such as Keypoint (Keypoint MRCNN) and multi-task learning (Hydra MRCNN) that have been developed to address the few-pixel feature limitations inherent in detecting objects such as cars in aerial datasets. Through comparative experiments on benchmark UAV datasets, we demonstrate the effectiveness of selected techniques in improving detection accuracy for small targets and assess the generalization performance. Our findings indicate that while Keypoint MRCNN enhances recall by incorporating scale-invariant structural cues, Hydra MRCNN excels at capturing fine-grained features through dynamic multi-branch learning. We further evaluate cross-dataset generalization and discuss how inconsistent labeling—such as the presence or absence of ignore regions—impacts performance. This work contributes insights into robust small object detection under real-world UAV conditions and highlights the need for tailored dataset design and adaptive architectures.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.002 | 0.005 |
| Meta-epidemiology (narrow) | 0.002 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.003 | 0.001 |
| Science and technology studies | 0.001 | 0.000 |
| Scholarly communication | 0.002 | 0.002 |
| Open science | 0.001 | 0.001 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.001 | 0.001 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".