《计算机应用》唯一官方网站 ›› 2022, Vol. 42 ›› Issue (7): 2058-2064.DOI: 10.11772/j.issn.1001-9081.2021050798
所属专题: 人工智能
收稿日期:
2021-05-17
修回日期:
2021-09-13
接受日期:
2021-09-22
发布日期:
2021-09-13
出版日期:
2022-07-10
通讯作者:
张雷
作者简介:
王震宇(1996—),男,江苏扬州人,硕士研究生,主要研究方向:深度学习、模式识别、自然语言处理基金资助:
Zhenyu WANG, Lei ZHANG(), Wenbin GAO, Weiming QUAN
Received:
2021-05-17
Revised:
2021-09-13
Accepted:
2021-09-22
Online:
2021-09-13
Published:
2022-07-10
Contact:
Lei ZHANG
About author:
WANG Zhenyu, born in 1996, M. S. candidate. His research interests include deep learning, pattern recognition, natural language processing.Supported by:
摘要:
为了解决基于传感器数据的运动识别问题,利用深度卷积神经网络(CNN)在公开的OPPORTUNITY传感器数据集上进行运动识别,提出了一种改进的渐进式神经网络架构搜索(PNAS)算法。首先,神经网络模型设计过程中不再依赖于合适拓扑结构的手动选择,而是通过PNAS算法来设计最优拓扑结构以最大化F1分数;其次,使用基于序列模型的优化(SMBO)策略,在该策略中将按照复杂度从低到高的顺序搜索结构空间,同时学习一个代理函数以引导对结构空间的搜索;最后,将搜索过程中表现最好的20个模型在OPPORTUNIT数据集上进行完全训练,并从中选出表现最好的模型作为搜索到的最优架构。通过这种方式搜索到的最优架构在OPPORTUNITY数据集上的F1分数达到了93.08%,与进化算法搜索到的最优架构及DeepConvLSTM相比分别提升了1.34%和1.73%,证明该方法能够改进以前手工设计的模型结构,且是可行有效的。
中图分类号:
王震宇, 张雷, 高文彬, 权威铭. 基于渐进式神经网络架构搜索的人体运动识别[J]. 计算机应用, 2022, 42(7): 2058-2064.
Zhenyu WANG, Lei ZHANG, Wenbin GAO, Weiming QUAN. Human activity recognition based on progressive neural architecture search[J]. Journal of Computer Applications, 2022, 42(7): 2058-2064.
数据类型 | 数目 |
---|---|
合计 | 21 144 |
无动作 | 15 152 |
开门1 | 348 |
开门2 | 370 |
关门1 | 327 |
关门2 | 352 |
打开冰箱 | 442 |
关闭冰箱 | 379 |
打开洗碗机 | 293 |
关闭洗碗机 | 266 |
打开抽屉1 | 204 |
关闭抽屉1 | 176 |
打开抽屉2 | 201 |
关闭抽屉2 | 163 |
打开抽屉3 | 242 |
关闭抽屉3 | 234 |
擦桌子 | 377 |
喝水 | 1 341 |
拨开关 | 277 |
表1 数据统计信息
Tab. 1 Data statistics
数据类型 | 数目 |
---|---|
合计 | 21 144 |
无动作 | 15 152 |
开门1 | 348 |
开门2 | 370 |
关门1 | 327 |
关门2 | 352 |
打开冰箱 | 442 |
关闭冰箱 | 379 |
打开洗碗机 | 293 |
关闭洗碗机 | 266 |
打开抽屉1 | 204 |
关闭抽屉1 | 176 |
打开抽屉2 | 201 |
关闭抽屉2 | 163 |
打开抽屉3 | 242 |
关闭抽屉3 | 234 |
擦桌子 | 377 |
喝水 | 1 341 |
拨开关 | 277 |
准确度 | 组合1 | 组合2 | 组合3 | |||
---|---|---|---|---|---|---|
操作1 | 操作2 | 操作1 | 操作2 | 操作1 | 操作2 | |
0.943 8 | 3×5-5×3 | 3×5-5×3 | 5×5-5×5 | 3×3-3×3 | 5×5-5×5 | 3×3-3×3 |
5×5-5×5 | 3×3-3×3 | |||||
0.943 2 | 3×5-5×3 | 3×3-3×3 | 5×5-5×5 | 5×5-5×5 | 5×5-5×5 | 3×3-3×3 |
5×5-5×5 | 2×6-6×2 | 5×5-5×5 | 3×3-3×3 | |||
0.942 6 | 1×3-3×1 | 3×3-3×3 | 5×5-5×5 | 5×5-5×5 | 5×5-5×5 | 3×3-3×3 |
5×5-5×5 | 2×6-6×2 | 5×5-5×5 | 3×3-3×3 | |||
0.942 0 | 3×5-5×3 | 5×5-5×5 | 5×5-5×5 | 3×3-3×3 | 5×5-5×5 | 3×3-3×3 |
5×5-5×5 | 1×7-7×1 | 5×5-5×5 | 3×3-3×3 | |||
0.941 5 | 3×5-5×3 | 5×5-5×5 | 3×5-5×3 | 5×5-5×5 | 3×5-5×3 | 5×5-5×5 |
3×5-5×3 | 1×7-7×1 | 3×5-5×3 | 5×5-5×5 |
表2 排名前五的模型的拓扑结构和准确度
Tab. 2 Topologies and accuracies of the top5 models
准确度 | 组合1 | 组合2 | 组合3 | |||
---|---|---|---|---|---|---|
操作1 | 操作2 | 操作1 | 操作2 | 操作1 | 操作2 | |
0.943 8 | 3×5-5×3 | 3×5-5×3 | 5×5-5×5 | 3×3-3×3 | 5×5-5×5 | 3×3-3×3 |
5×5-5×5 | 3×3-3×3 | |||||
0.943 2 | 3×5-5×3 | 3×3-3×3 | 5×5-5×5 | 5×5-5×5 | 5×5-5×5 | 3×3-3×3 |
5×5-5×5 | 2×6-6×2 | 5×5-5×5 | 3×3-3×3 | |||
0.942 6 | 1×3-3×1 | 3×3-3×3 | 5×5-5×5 | 5×5-5×5 | 5×5-5×5 | 3×3-3×3 |
5×5-5×5 | 2×6-6×2 | 5×5-5×5 | 3×3-3×3 | |||
0.942 0 | 3×5-5×3 | 5×5-5×5 | 5×5-5×5 | 3×3-3×3 | 5×5-5×5 | 3×3-3×3 |
5×5-5×5 | 1×7-7×1 | 5×5-5×5 | 3×3-3×3 | |||
0.941 5 | 3×5-5×3 | 5×5-5×5 | 3×5-5×3 | 5×5-5×5 | 3×5-5×3 | 5×5-5×5 |
3×5-5×3 | 1×7-7×1 | 3×5-5×3 | 5×5-5×5 |
编号 | 平均数 | 标准差 | 中位数 | 最小值 | 最大值 |
---|---|---|---|---|---|
1 | 0.919 522 | 0.005 528 | 0.920 868 | 0.902 169 | 0.927 428 |
2 | 0.920 135 | 0.006 067 | 0.922 891 | 0.900 702 | 0.925 996 |
3 | 0.924 570 | 0.004 411 | 0.926 382 | 0.906 374 | 0.929 054 |
4 | 0.901 550 | 0.026 100 | 0.908 650 | 0.797 012 | 0.921 621 |
5 | 0.917 748 | 0.009 842 | 0.920 067 | 0.866 850 | 0.924 272 |
6 | 0.920 104 | 0.005 025 | 0.922 175 | 0.901 746 | 0.925 013 |
7 | 0.919 362 | 0.006 375 | 0.921 375 | 0.897 056 | 0.924 669 |
8 | 0.919 620 | 0.005 268 | 0.921 226 | 0.895 362 | 0.924 658 |
9 | 0.919 226 | 0.009 095 | 0.922 838 | 0.880 309 | 0.924 767 |
10 | 0.922 278 | 0.004 002 | 0.922 978 | 0.903 349 | 0.927 468 |
11 | 0.919 174 | 0.014 942 | 0.924 615 | 0.862 486 | 0.926 538 |
12 | 0.920 456 | 0.004 927 | 0.921 955 | 0.903 347 | 0.927 389 |
13 | 0.926 368 | 0.005 139 | 0.928 377 | 0.909 028 | 0.930 858 |
14 | 0.920 531 | 0.012 185 | 0.924 319 | 0.864 361 | 0.928 656 |
15 | 0.922 617 | 0.004 133 | 0.924 509 | 0.909 749 | 0.926 209 |
16 | 0.923 063 | 0.002 901 | 0.923 541 | 0.908 954 | 0.925 753 |
17 | 0.902 136 | 0.021 373 | 0.909 083 | 0.832 309 | 0.925 316 |
18 | 0.926 029 | 0.004 275 | 0.927 353 | 0.911 680 | 0.930 610 |
19 | 0.922 370 | 0.007 389 | 0.924 482 | 0.893 722 | 0.926 957 |
20 | 0.921 048 | 0.002 975 | 0.921 664 | 0.910 051 | 0.925 131 |
表3 表现最好的20个模型的F1分数的统计信息
Tab. 3 Statistics of F1 scores of the best 20 models
编号 | 平均数 | 标准差 | 中位数 | 最小值 | 最大值 |
---|---|---|---|---|---|
1 | 0.919 522 | 0.005 528 | 0.920 868 | 0.902 169 | 0.927 428 |
2 | 0.920 135 | 0.006 067 | 0.922 891 | 0.900 702 | 0.925 996 |
3 | 0.924 570 | 0.004 411 | 0.926 382 | 0.906 374 | 0.929 054 |
4 | 0.901 550 | 0.026 100 | 0.908 650 | 0.797 012 | 0.921 621 |
5 | 0.917 748 | 0.009 842 | 0.920 067 | 0.866 850 | 0.924 272 |
6 | 0.920 104 | 0.005 025 | 0.922 175 | 0.901 746 | 0.925 013 |
7 | 0.919 362 | 0.006 375 | 0.921 375 | 0.897 056 | 0.924 669 |
8 | 0.919 620 | 0.005 268 | 0.921 226 | 0.895 362 | 0.924 658 |
9 | 0.919 226 | 0.009 095 | 0.922 838 | 0.880 309 | 0.924 767 |
10 | 0.922 278 | 0.004 002 | 0.922 978 | 0.903 349 | 0.927 468 |
11 | 0.919 174 | 0.014 942 | 0.924 615 | 0.862 486 | 0.926 538 |
12 | 0.920 456 | 0.004 927 | 0.921 955 | 0.903 347 | 0.927 389 |
13 | 0.926 368 | 0.005 139 | 0.928 377 | 0.909 028 | 0.930 858 |
14 | 0.920 531 | 0.012 185 | 0.924 319 | 0.864 361 | 0.928 656 |
15 | 0.922 617 | 0.004 133 | 0.924 509 | 0.909 749 | 0.926 209 |
16 | 0.923 063 | 0.002 901 | 0.923 541 | 0.908 954 | 0.925 753 |
17 | 0.902 136 | 0.021 373 | 0.909 083 | 0.832 309 | 0.925 316 |
18 | 0.926 029 | 0.004 275 | 0.927 353 | 0.911 680 | 0.930 610 |
19 | 0.922 370 | 0.007 389 | 0.924 482 | 0.893 722 | 0.926 957 |
20 | 0.921 048 | 0.002 975 | 0.921 664 | 0.910 051 | 0.925 131 |
模型 | 测试集F1得分 |
---|---|
CNN[ | 0.851 0 |
BaselineCNN[ | 0.883 0 |
DeepConvLSTM[ | 0.915 0 |
b-LSTM-S[ | 0.927 0 |
EA-Single-best[ | 0.918 5 |
EA-Ensemble-best[ | 0.927 5 |
PNASNet-best | 0.930 8 |
表4 不同模型的最优结果比较
Tab. 4 Comparison of optimal results of different models
模型 | 测试集F1得分 |
---|---|
CNN[ | 0.851 0 |
BaselineCNN[ | 0.883 0 |
DeepConvLSTM[ | 0.915 0 |
b-LSTM-S[ | 0.927 0 |
EA-Single-best[ | 0.918 5 |
EA-Ensemble-best[ | 0.927 5 |
PNASNet-best | 0.930 8 |
1 | 郑浦,白宏阳,李政茂,等. 抖动干扰下运动目标精准检测与跟踪算法设计[J]. 仪器仪表学报, 2019, 40(11):90-98. |
ZHENG P, BAI H Y, LI Z M, et al. Design of accurate detection and tracking algorithm for moving target under jitter interference[J]. Chinese Journal of Scientific Instrument, 2019, 40(11): 90-98. | |
2 | 吴天舒,陈蜀宇,吴朋. 全生命周期健康监测诊断系统研究[J]. 仪器仪表学报, 2018, 39(8):204-211. |
WU T S, CHEN S Y, WU P. Research on the life cycle health monitoring and diagnosis system[J]. Chinese Journal of Scientific Instrument, 2018, 39(8): 204-211. | |
3 | ALAA M, ZAIDAN A A, ZAIDAN B B, et al. A review of smart home applications based on internet of things[J]. Journal of Network and Computer Applications, 2017, 97: 48-65. 10.1016/j.jnca.2017.08.017 |
4 | CORNACCHIA M, OZCAN K, ZHENG Y, et al. A survey on activity detection and classification using wearable sensors[J]. IEEE Sensors Journal, 2017, 17(2): 386-403. 10.1109/jsen.2016.2628346 |
5 | ORDÓÑEZ F J, ROGGEN D. Deep convolutional and LSTM recurrent neural networks for multimodal wearable activity recognition[J]. Sensors, 2016, 16(1): No.115. 10.3390/s16010115 |
6 | HAMMERLA N Y, HALLORAN S, PLÖTZ T. Deep, convolutional, and recurrent models for human activity recognition using wearables[C]// Proceedings of the 25th International Joint Conference on Artificial Intelligence. California: IJCAI.org, 2016: 1533-1540. |
7 | WANG K, HE J, ZHANG L. Attention-based convolutional neural network for weakly labeled human activities’ recognition with wearable sensors[J]. IEEE Sensors Journal, 2019, 19(17): 7598-7604. 10.1109/jsen.2019.2917225 |
8 | TENG Q, WANG K, ZHANG L, et al. The layer-wise training convolutional neural networks using local loss for sensor-based human activity recognition[J]. IEEE Sensors Journal, 2020, 20(13): 7265-7274. 10.1109/jsen.2020.2978772 |
9 | 郑增威,杜俊杰,霍梅梅,等. 基于可穿戴传感器的人体活动识别研究综述[J]. 计算机应用, 2018, 38(5):1223-1229, 1238. 10.3390/s22051911 |
ZHENG Z W, DU J J, HUO M M, et al, Review of human activity recognition based on wearable sensors[J]. Journal of Computer Applications, 2018, 38(5): 1223-1229, 1238. 10.3390/s22051911 | |
10 | STANLEY K O, MIIKKULAINEN R. Evolving neural networks through augmenting topologies[J]. Evolutionary Computation, 2002, 10(2): 99-127. 10.1162/106365602320169811 |
11 | REAL E, MOORE S, SELLE A, et al. Large-scale evolution of image classifiers[C]// Proceedings of the 34th International Conference on Machine Learning. New York: JMLR.org, 2017: 2902-2911. |
12 | SUGANUMA M, SHIRAKAWA S, NAGAO T. A genetic programming approach to designing convolutional neural network architectures[C]// Proceedings of the 2017 Genetic and Evolutionary Computation Conference. New York: ACM, 2017: 497-504. 10.1145/3071178.3071229 |
13 | ELSKEN T, METZEN J H, HUTTER F. Efficient multi-objective neural architecture search via Lamarckian evolution[EB/OL]. (2019-02-26) [2021-05-17].. 10.1007/978-3-030-05318-5_3 |
14 | WILLIAMS R J. Simple statistical gradient-following algorithms for connectionist reinforcement learning[J]. Machine Learning, 1992, 8(3/4): 229-256. 10.1007/bf00992696 |
15 | BAKER B, GUPTA O, NAIK N, et al. Designing neural network architectures using reinforcement learning[EB/OL]. (2017-03-22) [2021-05-17].. |
16 | ZOPH B, LE QUOC V. Neural architecture search with reinforcement learning[EB/OL]. (2017-02-15) [2021-05-17].. |
17 | ZOPH B, VASUDEVAN V, SHLENS J, et al. Learning transferable architectures for scalable image recognition[C]// Proceedings of the 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2018: 8697-8710. 10.1109/cvpr.2018.00907 |
18 | LIU C X, ZOPH B, NEUMANN M, et al. Progressive neural architecture search[C]// Proceedings of the 2018 European Conference on Computer Vision, LNCS 11205. Cham: Springer, 2018: 19-35. |
19 | BALDOMINOS A, SAEZ Y, ISASI P. Evolutionary design of convolutional neural networks for human activity recognition in sensor-rich environments[J]. Sensors, 2018, 18(4): No.1288. 10.3390/s18041288 |
20 | NEGRINHO R, GORDON G. DeepArchitect: automatically designing and training deep architectures[EB/OL]. (2017-04-28) [2021-05-17].. |
21 | BROCK A, LIM T, RITCHIE J M, et al. SMASH: one-shot model architecture search through HyperNetworks[EB/OL]. (2017-08-17) [2021-05-17].. |
22 | DOMHAN T, SPRINGENBERG J T, HUTTER F. Speeding up automatic hyperparameter optimization of deep neural networks by extrapolation of learning curves[C]// Proceedings of the 24th International Joint Conference on Artificial Intelligence. Palo Alto, CA: AAAI Press, 2015: 3460-3468. |
23 | BAKER B, GUPTA O, RASKAR R, et al. Accelerating neural architecture search using performance prediction[EB/OL]. (2017-11-08) [2021-05-17].. |
24 | CHAVARRIAGA R, SAGHA H, CALATRONI A, et al. The opportunity challenge: a benchmark database for on-body sensor-based activity recognition[J]. Pattern Recognition Letters, 2013, 34(15): 2033-2042. 10.1016/j.patrec.2012.12.014 |
25 | YANG J B, NGUYEN M N, SAN P P, et al. Deep convolutional neural networks on multichannel time series for human activity recognition[C]// Proceedings of the 24th International Joint Conference on Artificial Intelligence. Palo Alto, CA: AAAI Press, 2015: 3995-4001. |
[1] | 潘烨新, 杨哲. 基于多级特征双向融合的小目标检测优化模型[J]. 《计算机应用》唯一官方网站, 2024, 44(9): 2871-2877. |
[2] | 李云, 王富铕, 井佩光, 王粟, 肖澳. 基于不确定度感知的帧关联短视频事件检测方法[J]. 《计算机应用》唯一官方网站, 2024, 44(9): 2903-2910. |
[3] | 黄云川, 江永全, 黄骏涛, 杨燕. 基于元图同构网络的分子毒性预测[J]. 《计算机应用》唯一官方网站, 2024, 44(9): 2964-2969. |
[4] | 秦璟, 秦志光, 李发礼, 彭悦恒. 基于概率稀疏自注意力神经网络的重性抑郁疾患诊断[J]. 《计算机应用》唯一官方网站, 2024, 44(9): 2970-2974. |
[5] | 王熙源, 张战成, 徐少康, 张宝成, 罗晓清, 胡伏原. 面向手术导航3D/2D配准的无监督跨域迁移网络[J]. 《计算机应用》唯一官方网站, 2024, 44(9): 2911-2918. |
[6] | 李顺勇, 李师毅, 胥瑞, 赵兴旺. 基于自注意力融合的不完整多视图聚类算法[J]. 《计算机应用》唯一官方网站, 2024, 44(9): 2696-2703. |
[7] | 陈虹, 齐兵, 金海波, 武聪, 张立昂. 融合1D-CNN与BiGRU的类不平衡流量异常检测[J]. 《计算机应用》唯一官方网站, 2024, 44(8): 2493-2499. |
[8] | 赵宇博, 张丽萍, 闫盛, 侯敏, 高茂. 基于改进分段卷积神经网络和知识蒸馏的学科知识实体间关系抽取[J]. 《计算机应用》唯一官方网站, 2024, 44(8): 2421-2429. |
[9] | 张春雪, 仇丽青, 孙承爱, 荆彩霞. 基于两阶段动态兴趣识别的购买行为预测模型[J]. 《计算机应用》唯一官方网站, 2024, 44(8): 2365-2371. |
[10] | 刘禹含, 吉根林, 张红苹. 基于骨架图与混合注意力的视频行人异常检测方法[J]. 《计算机应用》唯一官方网站, 2024, 44(8): 2551-2557. |
[11] | 顾焰杰, 张英俊, 刘晓倩, 周围, 孙威. 基于时空多图融合的交通流量预测[J]. 《计算机应用》唯一官方网站, 2024, 44(8): 2618-2625. |
[12] | 石乾宏, 杨燕, 江永全, 欧阳小草, 范武波, 陈强, 姜涛, 李媛. 面向空气质量预测的多粒度突变拟合网络[J]. 《计算机应用》唯一官方网站, 2024, 44(8): 2643-2650. |
[13] | 赵亦群, 张志禹, 董雪. 基于密集残差物理信息神经网络的各向异性旅行时计算方法[J]. 《计算机应用》唯一官方网站, 2024, 44(7): 2310-2318. |
[14] | 高阳峄, 雷涛, 杜晓刚, 李岁永, 王营博, 闵重丹. 基于像素距离图和四维动态卷积网络的密集人群计数与定位方法[J]. 《计算机应用》唯一官方网站, 2024, 44(7): 2233-2242. |
[15] | 徐松, 张文博, 王一帆. 基于时空信息的轻量视频显著性目标检测网络[J]. 《计算机应用》唯一官方网站, 2024, 44(7): 2192-2199. |
阅读次数 | ||||||
全文 |
|
|||||
摘要 |
|
|||||