小波时频局部化无人机目标检测模型压缩研究

doi:10.7527/S1000-6893.2025.31952

本期目录 | 过刊浏览 | 高级检索

前一篇 | 后一篇

小波时频局部化无人机目标检测模型压缩研究

黄维¹,潘家皓²,何楚¹

1. 湖北省武汉市武汉大学电子信息学院
2. 武汉大学

收稿日期:2025-03-10 修回日期:2025-05-06 出版日期:2025-05-08 发布日期:2025-05-08
通讯作者: 何楚
基金资助:
国家重点研发计划项目;国家自然科学基金项目

Wavelet Time-Frequency Localization-Based Model Compression for UAV Object Detection

¹,Jia-Hao PAN², ¹

1.
2. Wuhan University

Received:2025-03-10 Revised:2025-05-06 Online:2025-05-08 Published:2025-05-08

摘要/Abstract

摘要： 当前基于深度学习的遥感目标检测模型依托地面服务器强大的算力与存储资源，已具备高效处理海量遥感数据的能力。然而，在无人机等移动边缘设备部署场景中，受限于计算资源、存储容量与能耗约束，大规模模型难以实现有效部署。模型压缩，如轻量化设计、模型量化等，已成为了推动遥感目标检测算法落地的重要技术。本文以单阶段的目标检测深度网络为基准，提出一种基于小波时频局部化的模型压缩框架：（1）通过深度可分离卷积与离散小波变换的空频局部化特性相融合，构建具有扩展感受野的小波深度可分离卷积（W-DSConv）模块，实现了模型的轻量化重构；（2）基于小波频域分解特性，提出小波分频量化（W-FDQ）的量化感知训练方法，实现不同频段特征的独立量化，进一步完成对轻量化模型的压缩。实验中我们选取了YOLO系列的网络模型进行验证，在VisDrone2021无人机遥感数据集上的实验表明：（1）所提W-DSConv模块在参数量减少45.5%，计算量减少37.4%的情况下，检测精度波动幅度控制在2.2%以内；（2）采用W-FDQ方法进行6比特与4比特量化时，量化模型分别保持了浮点模型的95.8%与92.9%的检测性能。本研究为移动端遥感目标检测模型的轻量化部署提供了新的技术思路。

关键词: 目标检测, 无人机, 模型压缩, 轻量化, 模型量化, 小波变换

Abstract: Current deep learning-based remote sensing object detection models rely on the powerful computing and storage re-sources of ground servers, enabling efficient processing of massive remote sensing data. However, in deployment scenarios for mobile edge devices such as Unmanned Aerial Vehicles (UAVs), the limited computational resources, storage capacity, and energy constraints make it challenging to effectively deploy large-scale models. Model compression techniques, such as lightweight design and model quantization, have become critical for enabling the practical application of remote sensing object detection algorithms. This paper proposes a wavelet time-frequency localization-based model compression framework, using the single-stage object de-tection network as the baseline: (1) By integrating depthwise separable convolution with the spatial-frequency localization charac-teristics of discrete wavelet transform, a Wavelet Depthwise Separable Convolution (W-DSConv) module is constructed to achieve lightweight model reconstruction while expanding its receptive field; (2) Leveraging wavelet frequency domain decomposition, a Wavelet Frequency-Division Quantization (W-FDQ) method for quantization-aware training is proposed, enabling independent quantization of features across different frequency bands to further compress the lightweight model. Experiments are conducted using YOLO-series models on the VisDrone2021 UAV remote sensing dataset. Results demonstrate that: (1) The W-DSConv module reduces model parameters by 45.5% and computational load by 37.4%, while limiting detection accuracy fluctuations to within 2.2%; (2) When applying 6-bit and 4-bit quantization via W-FDQ, the quantized models retain 95.8% and 92.9% of the floating-point model’s detection performance, respectively. This research provides novel technical insights for lightweight deployment of remote sensing object detection models on mobile platforms.

Key words: Object Detection, Unmanned Aerial Vehicles, Model Compression, Lightweight, Model Quantization, Wavelet Transforms

中图分类号:

黄维潘家皓何楚. 小波时频局部化无人机目标检测模型压缩研究[J]. 航空学报, doi: 10.7527/S1000-6893.2025.31952.

Jia-Hao PAN. Wavelet Time-Frequency Localization-Based Model Compression for UAV Object Detection[J]. Acta Aeronautica et Astronautica Sinica, doi: 10.7527/S1000-6893.2025.31952.

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):524519-524519
[2]JIANG B, QU R K, LI Y D, et al.Object detection in UAV imagery based on deep learning: Review[J].ACTA AERONAUTICAET ASTRONAUTICA SINICA, 2021, 42(4):524519-524519
[3]欧阳权, 张怡, 马延, 等.基于深度学习的无人机航拍目标检测与跟踪方法综述[J].电光与控制, 2024, 31(3):1-7
[4]OUYANG Q, ZHANG Y, MA Y, et al.A Review of UAV Aerial Photography Target Detection and Tracking Methods Based on Deep Learning[J].Electronics Optics & Control, 2024, 31(3):1-7
[5]赵禄达, 胡以华, 赵楠翔, 等.点云深度学习模型的压缩和部署加速方法研究现状与展望特邀[J].激光与光电子学进展, 2024, 61(20):2011005-2011005
[6]ZHAO L D, HU Y H, ZHAO N X, et al.Review of Model Compression and Accelerated Development for Deep Learning in LiDAR Point Cloud Processing (Invited)[J].Laser & Optoelectronics Progress, 2024, 61(20):2011005-2011005
[7]CHEN F H, LI S L, HAN J L, et al.Review of light-weight deep convolutional neural networks[J].Archives of Computational Methods in Engineering, 2024, 31(4):1915-1937
[8]王军, 冯孙铖, 程勇.深度学习的轻量化神经网络结构研究综述[J].计算机工程, 2021, 47(8):1-13
[9]WANG J, FENG S C, CHENG Y.Survey of Research on Lightweight Neural Network Structures for Deep Learning[J].Computer Engineering, 2021, 47(8):1-13
[10]SIFRE L, MALLAT S. Rigid-motion scattering for tex-ture classification[J]. arXiv prep.[J].rXiv:1403.1687, 2014., rint, :-
[11]HOWARD A G, ZHU M L, CHEN B, et al. Mobilenets: Efficient convolutional neural networks for mobile vision applications[J]. arXiv prep.[J].rXiv:1704.04861, 2017., rint, :-
[12]SANDLER M, HOWARD A, ZHU M L, et al.Mo-bilenetv2: Inverted residuals and linear bottle-necks[C]//Proceedings of the IEEE conference on com-puter vision and pattern recognition. 2018: 4510-4520.
[13]ZHANG X Y, ZHOU X Y, LIN M X, et al.Shufflenet: An extremely efficient convolutional neural network for mobile devices[C]//Proceedings of the IEEE conference on computer vision and pattern recognition. 2018: 6848-6856.
[14]HAN K, WANG Y H, TIAN Q, et al.Ghostnet: More features from cheap operations[C]//Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 2020: 1580-1589.
[15]VASU P K A, GABRIEL J, ZHU J, et al.Mobileone: An improved one millisecond mobile back-bone[C]//Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 2023: 7907-7917.
[16]HAN S, MAO H Z, DALLY W J. Deep compression: Compressing deep neural networks with pruning, trained quantization and huffman coding[J]. arXiv prep.[J].rXiv:1510.00149, 2015., rint, :-
[17]LIU X C, YE M, ZHOU D Y, et al.Post-training quanti-zation with multiple points: Mixed precision without mixed precision[C]//Proceedings of the AAAI confer-ence on artificial intelligence. 2021, 35(10): 8697-8705.
[18]NAGEL M, AMJAD R A, VAN BAALEN M, et al.Up or down? adaptive rounding for post-training quantiza-tion[C]//International conference on machine learning. PMLR, 2020: 7197-7206.
[19]YUAN Z H, XUE C H, CHEN Y Q, et al.Ptq4vit: Post-training quantization for vision transformers with twin uniform quantization[C]//European conference on com-puter vision. Cham: Springer Nature Switzerland, 2022: 191-207.
[20]ESSER S K, MCKINSTRY J L, BABLANI D, et al. Learned step size quantization[J]. arXiv prep.[J].rXiv:1902.08153, 2019., rint, :-
[21]BHALGAT Y, LEE J, NAGEL M, et al.Lsq+: Improv-ing low-bit quantization through learnable offsets and better initialization[C]//Proceedings of the IEEE/CVF conference on computer vision and pattern recognition workshops. 2020: 696-697.
[22]CHOI J, WANG Z, VENKATARAMANI S, et al. Pact: Parameterized clipping activation for quantized neural networks[J]. arXiv prep.[J].rXiv:1805.06085, 2018., rint, :-
[23]LIU Z C, CHENG K T, HUANG D, et al.Nonuniform-to-uniform quantization: Towards accurate quantization via generalized straight-through estima-tion[C]//Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 2022: 4942-4952.
[24]ZHU K, HE Y Y, WU J X.Quantized feature distillation for network quantization[C]//Proceedings of the AAAI Conference on Artificial Intelligence. 2023, 37(9): 11452-11460.
[25]MUSA A, KAKUDI H A, HASSAN M, et al.Light-weight Deep Learning Models For Edge Devices—A Survey[J].International Journal of Computer Infor-mation Systems and Industrial Management Applications, 2025, 17:18-
[26]杨春, 张睿尧, 黄泷, 等.深度神经网络模型量化方法综述[J].工程科学学报, 2023, 45(10):1613-1629
[27]YANG C, ZHANG R Y, HUANG L, et al.A survey of quantization methods for deep neural net-works[J].Chinese Journal of Engineering, 2023, 45(10):1613-1629
[28]NAGEL M, FOURNARAKIS M, AMJAD R A, et al. A white paper on neural network quantization[J]. arXiv prep.[J].rXiv:2106.08295, 2021., rint, :-
[29]REDMON J, DIVVALA S, GIRSHICK R, et al.You only look once: Unified, real-time object detection[C] //2016 IEEE Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE Press, 2016: 779-788.
[30]REDMON J, FARHADI A. Yolov3: An incremental improvement[J]. arXiv prep.[J].rXiv:1804.02767, 2018., rint, :-
[31]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[C]//Proceedings of the IEEE/CVF conference on computer vision and pat-tern recognition. 2023: 7464-7475.
[32]FINDER S E, AMOYAL R, TREISTER E, et al.Wavelet convolutions for large receptive fields[C]//European Conference on Computer Vision. Cham: Springer Nature Switzerland, 2024: 363-380.
[33]PAN J H, HE C, HUANG W, et al.Wavelet Tree Trans-former: Multi-Head Attention with Frequency Selective Representation and Interaction for Remote Sensing Object Detection[J].IEEE Transactions on Geoscience and Remote Sensing, 2024, 62:1-23
[34]GONG R H, LIU X L, JIANG S H, et al.Differentiable soft quantization: Bridging full-precision and low-bit neural networks[C]//Proceedings of the IEEE/CVF inter-national conference on computer vision. 2019: 4852-4861.
[35]HUANG L, DONG Z W, CHEN S L, et al.HQOD: Harmonious Quantization for Object Detection[C]//2024 IEEE International Conference on Multimedia and Expo (ICME). IEEE, 2024: 1-6.

E-mail：hkxb@buaa.edu.cn

关于我们

期刊社服务

专业学科

封面文章

友情链接

主管单位：中国科学技术协会主办单位：中国航空学会北京航空航天大学

小波时频局部化无人机目标检测模型压缩研究

Wavelet Time-Frequency Localization-Based Model Compression for UAV Object Detection

PDF (PC)

可视化

摘要/Abstract

引用本文

使用本文

参考文献

相关文章 15

编辑推荐

Metrics

本文评价

[1]	向锦武, 马凯, 阚梓, 李道春, 郑可欣, 陈汉轩. 氢能源无人机关键技术研究进展[J]. 航空学报, 2025, 46(5): 531603-531603.
[2]	丁奇帅, 雷帮军, 吴正平. 基于孪生网络的轻量型无人机单目标跟踪算法[J]. 航空学报, 2025, 46(4): 330925-330925.
[3]	吴付杰, 王博文, 齐静雅, 曹铭智, 桑英俊, 李晟, 张玉珍, 陈钱, 左超. 机载多孔径全景图像合成技术研究进展[J]. 航空学报, 2025, 46(3): 630505-630505.
[4]	马诺, 卫社春, 孟军辉, 刘清洋, 雷宇声. 考虑减速伞作用的无人机内埋舱体分离流场特性与动力学[J]. 航空学报, 2025, 46(3): 130755-130755.
[5]	吴一全, 童康. 基于深度学习的无人机航拍图像小目标检测研究进展[J]. 航空学报, 2025, 46(3): 30848-030848.
[6]	郭廷宇, 闫溟, 解春雷. 聚合-分体飞行器气动特性[J]. 航空学报, 2024, 45(S1): 730596-730596.
[7]	余莎莎, 陈星雨. 城市空中交通领域关键技术创新与挑战[J]. 航空学报, 2024, 45(S1): 730657-730657.
[8]	马骥, 陈向勇, 温广辉, 程龙, 邱建龙. 随机切换拓扑下无人机集群预设时间跟踪控制[J]. 航空学报, 2024, 45(S1): 730793-730793.
[9]	柳汀, 周国鑫, 徐扬, 罗德林, 郭正玉, 杨梦杰. 融合信息图的优化哈里斯鹰多无人机动态目标搜索[J]. 航空学报, 2024, 45(S1): 730773-730773.
[10]	王海峰, 刘坤澎, 江泓鑫, 杜晨曦. 螺旋桨多设计点气动优化方法和变桨距角策略[J]. 航空学报, 2024, 45(9): 528831-528831.
[11]	赵春晖, 刘安萌, 吕洋, 潘泉. 无人机韧性自主定位技术综述[J]. 航空学报, 2024, 45(8): 28839-028839.
[12]	张薇, 何若俊. 面向物联网数据收集的无人机自主路径规划[J]. 航空学报, 2024, 45(8): 329054-329054-1.
[13]	赵靖, 宋丹. 无人机GNSS/IMU组合导航系统完好性监测方法[J]. 航空学报, 2024, 45(7): 328943-328943.
[14]	王传云, 苏阳, 王琳霖, 王田, 王静静, 高骞. 面向反制无人机集群的多目标连续鲁棒跟踪算法[J]. 航空学报, 2024, 45(7): 329017-329017.
[15]	罗旭东, 吴一全, 陈金林. 无人机航拍影像目标检测与语义分割的深度学习方法研究进展[J]. 航空学报, 2024, 45(6): 28822-028822.