YOLO-SWR：无人机视角下轻量级交通车辆检测算法

doi:10.3778/j.issn.1002-8331.2412-0303

摘要/Abstract

摘要： 在智能交通系统中，无人机视角的车辆检测因其灵活性和高效性，受到越来越多的关注。无人机的车辆检测面临目标尺寸小、尺度变化大、复杂背景干扰、计算资源有限等问题，如何在有限的计算资源下提升检测精度是一个重要的挑战。针对以上问题，提出了一种精度更高且轻量的车辆检测模型YOLO-SWR。在保持三尺度检测的前提下增加了针对小目标的检测层，充分提取小目标的位置信息和细节特征，同时采用共享轻量检测头，减少网络参数量和计算量；在骨干和颈部网络中使用小波变换替代传统卷积，利用低频和高频分量提取多尺度特征；在C3k2模块中集成RDS模块，加强从网络高层的可扩展感受野中提取特征，并通过残差结构和深度可分离卷积融合浅层特征和深层特征，同时缓解梯度消失问题；采用Soft-NMS策略优化目标框筛选过程，提升细节特征的处理精度。实验结果表明，基于筛选后的VisDrone2019数据集，提出的模型YOLO-SWR比基础模型YOLOv11n，在mAP方面提升了9.6个百分点，同时参数量减少了56%。因此，提出的模型YOLO-SWR在精度与轻量化方面均具有显著优势。

关键词: 小目标检测, 轻量化, 交通车辆, 目标跟踪, 无人机视角

Abstract: In intelligent transportation systems, vehicle detection from the perspective of unmanned aerial vehicle (UAV) has attracted more and more attention due to its flexibility and efficiency. Vehicle detection by UAVs faces problems such as small target size, large scale variation, complex background interference, and limited computational resources. How to improve the detection accuracy with limited computing resources is an important challenge. To solve the above problems, this paper proposes YOLO-SWR, a vehicle detection model with higher accuracy and lightweightness. Firstly, the model adds a detection layer for small objects, maintains the three-scale detection to fully extract the location information and detailed features of small objects, and adopts a shared lightweight detection head to reduce the number of network parameters and computational complexity. Secondly, wavelet pooling is used in the backbone and neck networks instead of traditional convolutions, leveraging both low-frequency and high-frequency components to improve multi-scale feature extraction. Furthermore, the RDS module is integrated into the C3k2 module to strengthen the extraction of features from the scalable receptive field at the high-level of the network, and the shallow features and deep features are fused through the residual structure and depthwise separable convolution, while alleviating the gradient disappearance problem. Finally, the Soft-NMS strategy is used to optimize the bounding box selection process to improve the processing precision of detailed features. Experimental results show that based on the filtered VisDrone2019 dataset, the proposed model YOLO-SWR improves mAP by 9.6 percentage points and reduces the number of parameters by 56%, compared with the base model YOLOv11n. Therefore, the YOLO-SWR model proposed in this paper has significant advantages in both accuracy and lightweightness.

Key words: small object detection, lightweight, traffic vehicle, object tracking, UAV perspective

王泉, 叶广飞, 陈祺东. YOLO-SWR：无人机视角下轻量级交通车辆检测算法[J]. 计算机工程与应用, 2025, 61(14): 112-122.

WANG Quan, YE Guangfei, CHEN Qidong. YOLO-SWR: Lightweight Traffic Vehicle Detection Algorithm from UAV Perspective[J]. Computer Engineering and Applications, 2025, 61(14): 112-122.

参考文献

[1] 张阳婷, 黄德启, 王东伟, 等. 基于深度学习的目标检测算法研究与应用综述[J]. 计算机工程与应用, 2023, 59(18): 1-13.
ZHANG Y T, HUANG D Q, WANG D W, et al. Review on research and application of deep learning-based target detection algorithms[J]. Computer Engineering and Applications, 2023, 59(18): 1-13.
[2] REN S, HE K, GIRSHICK R, et al. Faster R-CNN: towards real-time object detection with region proposal networks[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2017, 39(6): 1137-1149.
[3] REIS D, KUPEC J, HONG J, et al. Real-time flying object detection with YOLOv8[J]. arXiv:2305.00972, 2023.
[4] KHANAM R, HUSSAIN M. YOLOv11: an overview of the key architectural enhancements[J]. arXiv:2410.17725, 2024.
[5] LIU W, ANGUELOV D, ERHAN D, et al. SSD: single shot multibox detector[C]//Proceedings of the 14th European Conference on Computer Vision, 2016: 21-37.
[6] LIN T Y, GOYAL P, GIRSHICK R, et al. Focal loss for dense object detection[J]. arXiv:1708.02002, 2017.
[7] 史燕燕, 史殿习, 乔子腾, 等. 小样本目标检测研究综述[J]. 计算机学报, 2023, 46(8): 1753-1780.
SHI Y Y, SHI D X, QIAO Z T, et al. A survey on recent advances in few-shot object detection[J]. Chinese Journal of Computers, 2023, 46(8): 1753-1780.
[8] 童小钟, 魏俊宇, 苏绍璟, 等. 融合注意力和多尺度特征的典型水面小目标检测[J]. 仪器仪表学报, 2023, 44(1): 212-222.
TONG X Z, WEI J Y, SU S J, et al. Typical small target detection on water surfaces fusing attention and multi-scale features[J]. Chinese Journal of Scientific Instrument, 2023, 44(1): 212-222.
[9] YI H, LIU B, ZHAO B, et al. Small object detection algorithm based on improved YOLOv8 for remote sensing[J]. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2024, 17: 1734-1747.
[10] 马俊燕, 常亚楠. MFE-YOLOX: 无人机航拍下密集小目标检测算法[J]. 重庆邮电大学学报 (自然科学版), 2024, 36(1): 128-135.
MA J Y, CHANG Y N. MFE-YOLOX: dense small target detection algorithm under UAV aerial photography[J]. Journal of Chongqing University of Posts and Telecommunications (Natural Science Edition), 2024, 36(1): 128-135.
[11] LIU J, ZHANG J, NI Y, et al. Small-object detection in remote sensing images with super-resolution perception[J]. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2024, 17: 15721-15734.
[12] 张华卫, 张文飞, 蒋占军, 等. 引入上下文信息和Attention Gate的GUS-YOLO遥感目标检测算法[J]. 计算机科学与探索, 2024, 18(2): 453-464.
ZHANG H W, ZHANG W F, JIANG Z J, et al. GUS-YOLO remote sensing target detection algorithm introducing context information and attention gate[J]. Journal of Frontiers of?Computer?Science and Technology, 2024, 18(2): 453-464.
[13] HE L, ZHOU Y, LIU L, et al. Research and application of YOLOv11-based object segmentation in intelligent recognition at construction sites[J]. Buildings, 2024, 14(12): 3777.
[14] 陈佳, 章坚武, 张浙亮. 基于上下文信息与注意力特征的欺骗语音检测[J]. 电信科学, 2023, 39(2): 92-102.
CHEN J, ZHANG J W, ZHANG Z L. Spoof speech detection based on context information and attention feature[J]. Telecommunications Science, 2023, 39(2): 92-102.
[15] CHOLLET F. Xception: deep learning with depthwise separable convolutions[J]. arXiv:1610.02357, 2016.
[16] HAN S, DING H, HAN Z, et al. Head-dominant enhancement with local count for better human detection in crowds[J]. IEEE Transactions on Automation Science and Engineering, 2024, 22: 8794-8804.
[17] 周思捷, 刘天奇, 陈天华. 基于改进YOLOv5算法的水稻病害识别研究[J]. 中国农机化学报, 2024, 45(8): 246-253.
ZHOU S J, LIU T Q, CHEN T H. Research on rice disease recognition based on improved YOLOv5 algorithm[J]. Journal of Chinese Agricultural Mechanization, 2024, 45(8): 246-253.
[18] DU D, ZHU P, WEN L, et al. VisDrone-DET2019: the vision meets drone object detection in image challenge results[C]//Proceedings of the IEEE/CVF International Conference on Computer Vision Workshop, 2019: 213-226.
[19] WEN L, ZHU P, DU D, et al. VisDrone-MOT2019: the vision meets drone multiple object tracking challenge results[C]//Proceedings of the IEEE/CVF International Conference on Computer Vision Workshop, 2019: 189-198.
[20] DUAN K, BAI S, XIE L, et al. CenterNet: keypoint triplets for object detection[C]//Proceedings of the IEEE/CVF International Conference on Computer Vision, 2019: 6568-6577.
[21] WANG C Y, BOCHKOVSKIY A, LIAO H Y M. YOLOv7: trainable bag-of-freebies sets new state-of-the-art for real-time object detectors[J]. arXiv:2207.02696, 2022.
[22] ZHAO Y, LYU W, XU S, et al. DETRs beat YOLOs on real-time object detection[J]. arXiv:2304.08069, 2023.
[23] GE Z, LIU S, WANG F, et al. YOLOX: exceeding YOLO series in 2021[J]. arXiv:2107.08430, 2021.
[24] WANG C Y, YEH I H, LIAO H Y M. YOLOv9: learning what you want to learn using programmable gradient information[J]. arXiv:2402.13616, 2024.
[25] WANG A, CHEN H, LIU L, et al. YOLOv10: real-time end-to-end object detection[J]. arXiv:2405.14458, 2024.
[26] ZHANG Y, SUN P, JIANG Y, et al. ByteTrack: multi-object tracking by associating every detection box[J]. arXiv:2110. 06864, 2021.