基于多尺度特征融合的空对地密集小目标检测算法

doi:10.19678/j.issn.1000-3428.0252144

摘要/Abstract

摘要： 无人机（UAV）低空航拍技术在数据采集、安全监控以及地形测绘等多个领域中得到了广泛的应用，显著提高了作业效率。然而在无人机航拍目标检测中，检测目标往往存在尺寸较小、密集且分布不均等问题。为此，提出一种基于改进YOLOv8网络的目标检测算法TOD-YOLO。首先，在主干网络部分引入RFAConv下采样模块，在增加较少参数量的情况下为卷积核提供有效的注意力，使得网络在特征提取阶段聚焦于更小的目标。其次，提出特征融合模块CSPOKM以及一种新的特征融合路径，提高模型对小目标特征信息的提取能力，提升特征融合网络的性能。然后，基于深度可分离卷积（DWConv）和高效多尺度注意力（EMA）设计轻量级注意力检测头DWA-Head替代原有的检测头，在降低参数量的同时提升模型的目标检测精度。最后，提出可动态调节正负样本系数的损失函数DA-MPDIoU，为易漏检、难分类的小目标样本分配更高的权重，解决完整交并比（CIoU）在小目标密度大的复杂场景下容易产生误检和漏检的问题。实验结果表明，改进算法在VisDrone2019数据集上mAP@0.5与mAP@0.5:0.95指标较YOLOv8n分别提升9.6和6.8百分点，同时结合在DOTA数据集上的泛化实验，充分验证了该算法在小目标检测任务中具备显著优势与巨大潜力。

关键词: 小目标检测, 无人机, 特征融合, DWA-Head, DA-MPDIoU

Abstract: Unmanned Aerial Vehicle (UAV) low-altitude aerial photography technology has been widely used in various fields, such as data acquisition, security monitoring, and terrain mapping, and has significantly improved operational efficiency. However, because aerial images typically cover a wide area, the detection targets are often small, dense, and unevenly distributed, limiting the accuracy of target detection. To address these issues, a target detection algorithm, TOD-YOLO, is proposed based on an improved YOLOv8 network. First, an RFAConv downsampling module is introduced into the backbone network, which provides effective attention to the convolutional kernel without increasing the number of parameters, thereby allowing the network to focus on smaller targets in the feature extraction phase. Second, the feature fusion module CSPOKM and a new feature fusion path are proposed to increase the focus on small targets and improve the performance of the feature fusion network. Subsequently, based on the depthwise separable convolution DWConv and the Efficient Multi-scale Attention (EMA) mechanism, the lightweight attention detection head DWA-Head is designed to reduce the number of parameters while improving the detection accuracy. Finally, to solve the problem of Complete Intersection over Union (CIoU) being easy to misdetect and remain undetected in complex scenes with a high density of small targets, DA-MPDIoU, a loss function that can dynamically adjust the positive and negative sample coefficients, is designed to assign higher weights to small target samples that are easy to miss and difficult to classify and optimize the training results. The experimental results show that, compared with the YOLOv8n algorithm, the proposed algorithm improves mAP@0.5 and mAP@0.5:0.95 on the VisDrone2019 dataset by 9.6 and 6.8 percentage points, respectively. Further generalization experiments are performed on the DOTA dataset. The experimental results show that the proposed algorithm exhibits significant advantages and potential for small-target detection tasks.

Key words: small target detection, Unmanned Aerial Vehicle (UAV), feature fusion, DWA-Head, DA-MPDIoU

中图分类号:

TP391

邓宇辉, 邓月明, 何鑫. 基于多尺度特征融合的空对地密集小目标检测算法[J]. 计算机工程, 2026, 52(11): 191-203.

DENG Yuhui, DENG Yueming, HE Xin. Air-to-Ground Dense Small Target Detection Algorithm Based on Multi-scale Feature Fusion[J]. Computer Engineering, 2026, 52(11): 191-203.

https://www.ecice06.com/CN/Y2026/V52/I11/191

参考文献

[1] 江波, 屈若锟, 李彦冬, 等. 基于深度学习的无人机航拍目标检测研究综述[J]. 航空学报, 2021, 42(4): 131-145. JIANG B, QU R K, LI Y D, et al. Object detection in UAV imagery based on deep learning: review[J]. Acta Aeronautica et Astronautica Sinica, 2021, 42(4): 131-145. (in Chinese)
[2] REDMON J, DIVVALA S, GIRSHICK R, et al. You only look once: unified, real-time object detection[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Las Vegas, USA: IEEE Press, 2016: 779-788.
[3] JOCHER G. YOLOv5 by Ultralytics (version 7.0)[EB/OL]. [2025-02-01]. https://doi.org/10.5281/zenodo.3908559.
[4] BOCHKOVSKIY A, WANG C Y, LIAO H M. YOLOv4: optimal speed and accuracy of object detection[EB/OL]. [2025-02-01]. https://arxiv.org/abs/2004.10934.
[5] GE Z, LIU S T, WANG F, et al. YOLOX: exceeding YOLO series in 2021[EB/OL]. [2025-02-01]. https://arxiv.org/abs/2107.08430.
[6] 王国明, 贾代旺. 基于YOLOv8的小目标检测模型的优化[J]. 计算机工程, 2025, 51(12): 294-303.WANG G M, JIA D W. Optimization of small object detection model based on YOLOv8[J]. Computer Engineering, 2025, 51(12): 294-303. (in Chinese)
[7] LI Y T, FAN Q S, HUANG H S, et al. A modified YOLOv8 detection network for UAV aerial image recognition[J]. Drones, 2023, 7(5): 304.
[8] TAN M X, PANG R M, LE Q V. EfficientDet: scalable and efficient object detection[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Seattle, USA: IEEE Press, 2020: 10778-10787.
[9] 程擎, 何汶键, 贾志杰, 等. 基于改进YOLOv7的低空小型无人机识别算法研究[J]. 贵州师范大学学报(自然科学版), 2025, 43(5): 45-53, 103.CHENG Q, HE W J, JIA Z J, et al. Research on low altitude small UAVs recognition algorithm based on improved YOLOv7[J]. Journal of Guizhou Normal University (Natural Sciences), 2025, 43(5): 45-53, 103. (in Chinese)
[10] WANG G, CHEN Y F, AN P, et al. UAV-YOLOv8: a small-object-detection model based on improved YOLOv8 for UAV aerial photography scenarios[J]. Sensors, 2023, 23(16): 7190.
[11] TONG Z J, CHEN Y H, XU Z W, et al. Wise-IoU: bounding box regression loss with dynamic focusing mechanism[EB/OL]. [2025-02-01]. https://arxiv.org/abs/2301.10051.
[12] 李子轩, 赵志刚, 张泽宇, 等. 基于FNB-YOLOv5的钢筋网绑扎点目标检测[J]. 上海交通大学学报, 2026, 60(5): 762-775.LI Z X, ZHAO Z G, ZHANG Z Y, et al. Object detection of steel mesh binding point using FNB-YOLOv5[J]. Journal of Shanghai Jiaotong University, 2026, 60(5): 762-775. (in Chinese)
[13] SUN Y H, LAN Z P, SUN Y G, et al. LDSTD: low-altitude drone aerial small target detector[J]. The Journal of Supercomputing, 2025, 81(2): 414.
[14] 廖宁生, 曹天秀, 刘科言, 等. 复合特征与多尺度融合的无人机小目标检测算法[J]. 计算机工程与应用, 2025, 61(3): 111-120.LIAO N S, CAO T X, LIU K Y, er al. Small target detection algorithm for UAV based on composite feature and multi-scale fusion[J]. Computer Engineering and Applications, 2025, 61(3): 111-120. (in Chinese)
[15] ZHANG X, LIU C, YANG D G, et al. RFAConv: receptive-field attention convolution for improving convolutional neural networks[EB/OL]. [2025-02-01]. https://arxiv.org/abs/2304.03198.
[16] SUNKARA R, LUO T. No more strided convolutions or pooling: a new CNN building block for low-resolution images and small objects[C]//Proceedings of Joint European Conference on Machine Learning and Knowledge Discovery in Databases. Berlin, Germany: Springer, 2022: 443-459.
[17] HOU Q B, ZHOU D Q, FENG J S. Coordinate attention for efficient mobile network design[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Nashville, USA: IEEE Press, 2021: 13708-13717.
[18] CUI Y N, REN W Q, KNOLL A. Omni-Kernel network for image restoration[C]//Proceedings of the AAAI Conference on Artificial Intelligence. [S. l]: AAAI Press, 2024: 1426-1434.
[19] ZHANG H Y, WANG Y, DAYOUB F, et al. VarifocalNet: an IoU-aware dense object detector[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Nashville, USA: IEEE Press, 2021: 8510-8519.
[20] YANG Y, LI M, MENG B, et al. Rethinking the aligned and misaligned features in onestage object detection[EB/OL]. [2025-02-01]. https://arxiv.org/abs/2108.12176.
[21] HOWARD A G, ZHU M L, CHEN B, et al. MobileNets: efficient convolutional neural networks for mobile vision applications[EB/OL]. [2025-02-01]. https://arxiv.org/abs/1704.04861.
[22] JOCHER G, QIU J, CHAURASIA A. Ultralytics YOLO (version 8.0.0)[EB/OL]. [2025-02-01]. https://github.com/ultralytics/ultralytics.
[23] OUYANG D L, HE S, ZHANG G Z, et al. Efficient multi-scale attention module with cross-spatial learning[C]//Proceedings of 2023 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). Rhodes Island, Greece: IEEE Press, 2023: 1-5.
[24] ZHENG Z H, WANG P, LIU W, et al. Distance-IoU loss: faster and better learning for bounding box regression[C]//Proceedings of the AAAI Conference on Artificial Intelligence. [S. l.]: AAAI Press, 2020: 12993-13000.
[25] MA S L, XU Y. MPDIoU: a loss for efficient and accurate bounding box regression[EB/OL]. [2025-02-01]. https://arxiv.org/abs/2307.07662.
[26] YU Z P, HUANG H B, CHEN W J, et al. YOLO-FaceV2: a scale and occlusion aware face detector[J]. Pattern Recognition, 2024, 155: 110714.
[27] CAO Y R, HE Z J, WANG L J, et al. VisDrone-DET2021: the vision meets drone object detection challenge results[C]//Proceedings of the IEEE/CVF International Conference on Computer Vision Workshops (ICCVW). Montreal, Canada: IEEE Press, 2021: 2847-2854.
[28] XIA G S, BAI X, DING J, et al. DOTA: a large-scale dataset for object detection in aerial images[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition （CVPR）. Salt Lake City, USA: IEEE Press, 2018: 3974-3983.
[29] YANG G Y, LEI J, ZHU Z K, et al. AFPN: asymptotic feature pyramid network for object detection[C]//Proceedings of the IEEE International Conference on Systems, Man, and Cybernetics (SMC). Honolulu, USA: IEEE Press, 2024: 2184-2189.
[30] 张博文, 薛波. 基于多尺度特征的无人机目标识别算法[J]. 电讯技术, 2025, 65(11): 1773-1780.ZHANG B W, XUE B. UAV target recognition algorithm based on multi-scale features[J]. Telecommunication Engineering, 2025, 65(11): 1773-1780. (in Chinese)
[31] LYU C Q, ZHANG W W, HUANG H A, et al. RTMDet: an empirical study of designing real-time object detectors[EB/OL]. [2025-02-01]. https://arxiv.org/abs/2212.07784.
[32] YANG C, HUANG Z H, WANG N Y. QueryDet: cascaded sparse query for accelerating high-resolution small object detection[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). New Orleans, USA: IEEE Press, 2022: 13658-13667.
[33] LIN T Y, MAIRE M, BELONGIE S, et al. Microsoft COCO: common objects in context[C]//Proceedings of the 13th European Conference on Computer Vision. Zurich, Switzerland: Springer, 2014: 740-755.
[34] WANG J W, YANG W, GUO H W, et al. Tiny object detection in aerial images[C]//Proceedings of the 25th International Conference on Pattern Recognition (ICPR). Milan, Italy: IEEE Press, 2021: 3791-3798.

选择文件类型/文献管理软件名称

选择包含的内容