Compiler Optimization for On-device Deep Learning in UAV Cooperative Localization

doi:10.3778/j.issn.1673-9418.2404077

Abstract

Abstract: With the rapid development of UAV technology, it has become a difficult research problem to locate UAV in the absence of positioning signal. In recent years, the emergence and development of graph neural network provides a new way to solve this problem. However, the deployment of image neural networks in resource-constrained UAVs faces challenges such as limited storage resources and difficulties in meeting real-time requirements. Under this background, an airborne depth calculation and compilation optimization method for UAV collaborative positioning is proposed in this paper. Specifically, this paper uses a lightweight time graph convolutional neural network model, which consists of a graph convolutional network and a gated recursive unit. The spatial dependence of UAV swarm and the time dependence of UAV position change are considered at the same time, and the position of UAV swarm is accurately predicted. Aiming at the redundancy of the model in the time-graph convolutional network, a global adaptive pruning algorithm based on inverse Cuthill-McKee graph rearrangement and double-depth deterministic strategy gradient is proposed. While ensuring the accurate prediction of UAV group coordinates, it can not only improve the spatial locality of data in main memory, accelerate the computing speed of the model, but also carry out adaptive unstructured pruning on the model and reduce the storage complexity of the model. Experimental results show that compared with the existing time graph convolutional neural network model, the compiled and optimized lightweight time graph convolutional neural network model retains 78.8% accuracy, while the calculation time of the model is reduced by 37.9%, and the average pruning rate of the model reaches 90.3%.

Key words: temporal graph convolutional network, cooperative localization, channel pruning, graph rearrangement algorithm, deep deterministic policy gradient

摘要： 随着无人机技术快速发展，在定位信号缺失的情况下进行无人机定位成为一个研究难题。而近几年图神经网络的出现与发展，为解决这一难题提供了一种新的解决思路。然而在资源受限的无人机端侧部署图神经网络面临着无人机算储资源受限及实时性难以满足等挑战。提出面向无人机协同定位的机载深度计算编译优化方法。采用了一种轻量化的时间图卷积神经网络模型，该时间图卷积网络由图卷积网络和门控递归单元组成，将无人机群的空间依赖性和无人机位置变化的时间依赖性同时加以考虑，对无人机群位置进行精确的预测；针对该模型在时间图卷积网络上的冗余特性，提出了基于逆向Cuthill-McKee图重排和基于双深度确定性策略梯度的全局自适应剪枝算法。在保证无人机群坐标精确预测的同时，不仅能提高数据在主存的空间局部性，加速模型的运算速度，而且能够对模型进行自适应的非结构化剪枝，降低模型的存储复杂度。实验结果表明，相对于已有的时间图卷积神经网络模型，编译优化后的轻量化时间图卷积神经网络模型在保留78.8%准确率的同时，模型计算时间降低37.9%，模型的平均剪枝率达到90.3%。

关键词: 时间图卷积网络, 协同定位, 通道剪枝, 图重排算法, 深度确定性策略梯度

XIONG Kang, LIU Sicong, WANG Hongtao, GAO Yuan, GUO Bin, YU Zhiwen. Compiler Optimization for On-device Deep Learning in UAV Cooperative Localization[J]. Journal of Frontiers of Computer Science and Technology, 2025, 19(1): 141-157.

熊康, 刘思聪, 王宏涛, 高元, 郭斌, 於志文. 面向无人机协同定位的机载深度计算编译优化[J]. 计算机科学与探索, 2025, 19(1): 141-157.

References

[1] GUPTA L, JAIN R, VASZKUN G. Survey of important issues in UAV communication networks[J] IEEE Communications Surveys & Tutorials, 2016, 18(2): 1123-1152.
[2] HAYAT S, YANMAZ E, MUZAFFAR R. Survey on unmanned aerial vehicle networks for civil applications: a communications viewpoint[J] IEEE Communications Surveys & Tutorials, 2016, 18(4): 2624-2661.
[3] YANG L, ZHANG W. Beam tracking and optimization for UAV communication[J]. IEEE Transactions on Wireless Communications, 2019, 18(11): 5367-5379.
[4] ZHANG C, ZHANG W, WANG W, et al. Research challenges and opportunities of UAV millimeter-wave communications[J]. IEEE Wireless Communications, 2019, 26(1): 58-62.
[5] ANDRE T, HUMMEL K A, SCHOELLIG A P, et al. Applica-tion-driven design of aerial communication networks[J]. IEEE Communications Magazine, 2014, 52(5): 129-137.
[6] MAO G, DRAKE S, ANDERSON B D O. Design of an extended Kalman filter for UAV localization[C]//Proceedings of the 2007 Information, Decision and Control. Piscataway: IEEE, 2007.
[7] KIM E, LEE S,KIM C, et al. Mobile beacon-based 3D-localization with multidimensional scaling in large sensor networks[J]. IEEE Communications Letters, 2010, 14(7): 647-649.
[8] SSU K F, OU C H, JIAU H C. Localization with mobile anchor points in wireless sensor networks[J]. IEEE Transactions on Vehicular Technology, 2005, 54(3): 1187-1197.
[9] VILLAS L A, GUIDONI D L, UEYAMA J. 3D localization in wireless sensor networks using unmanned aerial vehicle[C]//Proceedings of the 2013 IEEE International Symposium on Network Computing & Applications. Piscataway: IEEE, 2013: 135-142.
[10] YU T, WANG X, JIN J, et al. Cloud-orchestrated physical topology discovery of large-scale IoT systems using UAVs[J]. IEEE Transactions on Industrial Informatics, 2018, 14(5): 2261-2270.
[11] DENG L, LI J, HUANG J T, et al. Recent advances in deep learning for speech research at Microsoft[C]//Proceedings of the 2013 IEEE International Conference on Acoustics, Speech and Signal Processing. Piscataway: IEEE, 2013: 8604-8608.
[12] KRIZHEVSKY A, SUTSKEVER I, HINTON G E. ImageNet classification with deep convolutional neural networks[C]//Advances in Neural Information Processing Systems 25, Lake Tahoe, Dec 3-6, 2012: 1097-1105.
[13] COLLOBERT R, WESTON J, BOTTOU L, et al. Natural language processing (almost) from scratch[J]. Journal of Machine Learning Research, 2011, 12: 2493-2537.
[14] CHEN C, SEFF A, KORNHAUSER A, et al. Deep driving: learning affordance for direct perception in autonomous driving[C]//Proceedings of the 2015 IEEE International conference on Computer Vision. Washington: IEEE Computer Society, 2015: 2722-2730.
[15] ESTEVA A, KUPREL B, NOVOA R A, et al. Dermatologist-level classification of skin cancer with deep neural networks[J]. Nature, 2017, 542(7639): 115-118.
[16] SILVER D, HUANG A, MADDISON C J, et al. Mastering the game of Go with deep neural networks and tree search[J]. Nature, 2016, 529(7587): 484-489.
[17] GALLINARI P, THIRIA S, FOGELMAN F. Multilayer perceptrons and data analysis[C]//Proceedings of the IEEE 1988 International Conference on Neural Networks. Piscataway: IEEE, 1988: 391-399.
[18] LECUN Y, BOSER B, DENKER J S, et al. Backpropagation applied to handwritten zip code recognition[J]. Neural Computation, 1989, 1(4): 541-551.
[19] ELMAN J L. Finding structure in time[J]. Cognitive Science, 1990, 14(2): 179-211.
[20] GOODFELLOW I J, POUGET-ABADIE J, MIRZA M, et al. Generative adversarial networks[C]//Advances in Neural Information Processing Systems 27, Montreal, Dec 8-13, 2014: 2672-2680.
[21] RIFAI S, VINCENT P, MULLER X, et al. Contractive auto-encoders: explicit invariance during feature extraction[C]//Proceedings of the 28th International Conference on Machine Learning. Madison: Omnipress, 2011: 833-840.
[22] ZHANG L L, HAN S, WEI J, et al. nn-Meter: towards accurate latency prediction of deep-learning model inference on diverse edge devices[C]//Proceedings of the 19th Annual International Conference on Mobile Systems, Applications, and Services. New York: ACM, 2021: 81-93.
[23] LI Y, BAIK J, RAHMAN M M, et al. Pareto optimization of CNN models via hardware-aware neural architecture search for drainage crossing classification on resource-limited devices[C]//Proceedings of the 2023 Workshops of the International Conference on High Performance Computing, Network, Storage, and Analysis. New York: ACM, 2023: 1767-1775.
[24] NAIR S, ABBASI S, WONG A, et al. MAPLE-Edge: a runtime latency predictor for edge devices[C]//Proceedings of the 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2022: 3660-3668.
[25] KONG Y, YANG P, CHENG Y. Edge-assisted on-device model update for video analytics in adverse environments[C]//Proceedings of the 31st ACM International Conference on Multimedia. New York: ACM, 2023: 9051-9060.
[26] LIU S, WANG T, LI J, et al. AdaMask: enabling machine-centric video streaming with adaptive frame masking for DNN inference offloading[C]//Proceedings of the 30th ACM International Conference on Multimedia. New York: ACM, 2022: 3035-3044.
[27] 周飞燕, 金林鹏, 董军. 卷积神经网络研究综述[J]. 计算机学报, 2017, 40(6): 1229-1251.
ZHOU F Y, JIN L P, DONG J. Review of convolutional network[J]. Chinese Journal of Computers, 2017, 40(6): 1229-1251.
[28] GORI M, MONFARDINI G, SCARSELLI F. A new model for learning in graph domains[C]//Proceedings of the 2005 IEEE International Joint Conference on Neural Networks. Piscataway: IEEE, 2005: 729-734.
[29] BRUNA J, ZAREMBA W, SZLAM A, et al. Spectral networks and locally connected networks on graphs[C]//Proceedings of the 2nd International Conference on Learning Representations, Banff, Apr 14-16, 2014.
[30] VASWANI A, SHAZEER N, PARMAR N, et al. Attention is all you need[C]//Advances in Neural Information Processing Systems 30, Long Beach, Dec 4-9, 2017: 5998-6008.
[31] KIPF T N, WELLING M. Variational graph auto-encoders[EB/OL]. [2024-01-15]. https://arxiv.org/abs/1611.07308.
[32] PAN S, HU R, LONG G, et al. Adversarially regularized graph autoencoder for graph embedding[C]//Proceedings of the 27th International Joint Conference on Artificial Intelligence, Stockholm, Jul 13-19, 2018: 2609-2615.
[33] JOHNSON D D. Learning graphical state transitions[C]// Proceedings of the 5th International Conference on Learning Representations, Toulon, Apr 24-26, 2017.
[34] XU D, CHENG W, LUO D, et al. Spatio-temporal attentive RNN for node classification in temporal attributed graphs[C]//Proceedings of the 28th International Joint Conference on Artificial Intelligence, Macao, China, Aug 10-16, 2019: 3947-3953.
[35] LUO J H, WU J, LIN W. ThiNet: a filter level pruning method for deep neural network compression[C]//Proceedings of the 2017 IEEE International Conference on Computer Vision. Washington: IEEE Computer Society, 2017: 5058-5066.
[36] GUPTA S, AGRAWAL A, GOPALAKRISHNAN K, et al. Deep learning with limited numerical precision[C]//Proceedings of the 32nd International Conference on Machine Learning, Lille, Jul 6-11, 2015: 1737-1746.
[37] VANHOUCKE V, SENIOR A, MAO M Z. Improving the speed of neural networks on CPUs[C]//Advances in Neural Information Processing Systems 24, Granada, Dec 12-14, 2011: 4.
[38] ZAGORUYKO S, KOMODAKIS N. Paying more attention to attention: improving the performance of convolutional neural networks via attention transfer[EB/OL].[2024-01-15]. https://arxiv.org/abs/1612.03928.
[39] DEFFERRARD M, BRESSON X, VANDERGHEYNST P. Convolutional neural networks on graphs with fast localized spectral filtering[C]//Advances in Neural Information Processing Systems 29, Barcelona, Dec 5-10, 2016: 3837-3845.
[40] BENGIO Y, SIMARD P, FRASCONI P. Learning long-term dependencies with gradient descent is difficult[J]. IEEE Transactions on Neural Networks, 1994, 5(2): 157-166.
[41] HOCHREITER S, SCHMIDHUBER J. Long short-term memory[J]. Neural Computation, 1997, 9: 1735-1780.
[42] CHO K, VAN MERRIENBOER B, BAHDANAU D, et al. On the properties of neural machine translation: encoder-decoder approaches[EB/OL]. [2024-01-15]. https://arxiv.org/abs/1409.1259.
[43] CHUNG J, GULCEHRE C, CHO K, et al. Empirical evaluation of gated recurrent neural networks on sequence modeling[EB/OL]. [2024-01-15]. https://arxiv.org/abs/1412.3555.
[44] 苗旭鹏, 王驭捷, 沈佳, 等. 面向多GPU的图神经网络训练加速[J]. 软件学报, 2023, 34(9): 4407-4420.
MIAO X P, WANG Y J, SHEN J, et al. Graph neural network training acceleration for multi-GPUs[J]. Journal of Software, 2023, 34(9): 4407-4420.
[45] AZAD A, JACQUELIN M, BULUÇ A, et al. The reverse Cuthill-mcKee algorithm in distributed-memory[C]//Proceedings of the 2017 IEEE International Parallel and Distributed Processing Symposium. Washington: IEEE Computer Society, 2017: 22-31.
[46] CHIN T W, DING R Z, ZHANG C, et al. Towards efficient model compression via learned global ranking[C]//Proceedings of the 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2020: 1515-1525.
[47] 王炯. 基于结构化剪枝的深度神经网络模型压缩方法研究[D]. 南京: 南京邮电大学, 2022.
WANG J. Research on deep neural network model compression method based on structural pruning[D]. Nanjing: Nanjing University of Posts and Telecommunications, 2022.