-
摘要: 提出一種基于近鄰成分分析(Neighbourhood component analysis, NCA)、L2正則化和隨機配置網絡(Stochastic configuration networks, SCNs)的輕量型人體行為識別學習模型. 首先, 針對人體行為特征集維數過高且可分性差的問題, 利用NCA從特征集中選擇高相關性特征子集, 進而提高模型建模計算過程的輕量性和識別精度. 其次, 針對SCNs隱含層節點過多時容易出現過擬合的問題, 采用L2正則化方法增強SCNs的泛化能力, 同時利用監督機制約束產生隱含層參數的方法, 極大地提高了SCNs模型的輕量性. 最后, 將所提NCA?L2?SCNs學習模型在UCI HAR特征集上進行驗證, 實驗結果表明, 相比于其他模型, 本文所提輕量型模型對于人體行為識別具有更好的識別精度和更快的建模速度.Abstract: In the past few decades, smartphone-based human activity recognition research has played an important role in many fields, including smart buildings, healthcare, and the military. However, the CPU and storage space of smartphones are very limited, so developing a lightweight human activity recognition learning model has become a research focus and hot spot in this field. To address the abovementioned problems, this paper proposed a lightweight human activity recognition learning model based on the nearest neighbor component analysis (NCA), L2 regularization, and stochastic configuration networks (SCNs). In the proposed model, aiming first at the problem of high dimension and poor separability exhibited by the human activity data, NCA was used to select a subset of highly relevant data from the dataset to improve the lightness of calculation using the learning algorithm in the modeling process and recognition accuracy of the established model. Second, to prevent the occurrence of overfitting when there are too many hidden layer nodes in SCNs, the L2 regularization method was adopted to enhance the generalization ability of SCNs. At the same time, the method of using the supervision mechanism to restrict the generation of hidden layer parameters greatly improved the lightness of the SCNs model. Finally, the proposed learning model and other learning models were verified experimentally on the UCI human activity recognition dataset. Experimental results show that compared with SCNs, the proposed L2?SCNs model reduces the lightness of the number of parameters by 20% and helps improve the accuracy of the model. The introduction of the NCA method has greatly facilitated the recognition accuracy and lightness (modeling time) of the L2?SCNs model, increasing by 3.41% and 70.24%, respectively. Moreover, compared with other state-of-the-art models, such as the support vector machine and long short-term memory network, the proposed model achieves the best recognition accuracy of 97.48% in the shortest time. To sum up, the model proposed herein is a lightweight human activity recognition model with exceptional recognition accuracy and a fast modeling speed.
-
圖 2 使用L2正則化前后SCNs的RMSE對比圖
Figure 2. RMSE comparison of SCNs before and after L2 regularization
表 1 不同算法的計算復雜度對比
Table 1. Comparison of the computational complexity of different algorithms
Algorithms Computational complexity SCNs $O( {{m^3} + 2 \times 561 \times m} )$ L2?SCNs $ O( {{2 \mathord{\left/ {\vphantom {2 3}} \right. } 3}({D^3}) + 2 \times 561 \times m} ) $ NCA?L2?SCNs $ O( {{2 \mathord{\left/ {\vphantom {2 3}} \right. } 3}({D^3}) + 2 \times 111 \times m} ) $ 
下載: 導出CSV
表 2 使用NCA特征選擇前后L2?SCNs模型結果對比
Table 2. Comparison of L2?SCNs model results before and after using the NCA feature selection
Feature dimension Modeling time/s Average accuracy/% Minimum accuracy/% Maximum accuracy/% UCI feature set(561) 29.85 94.27 94.10 94.30 NCA(111) 18.03 97.48 97.05 97.93 NCA(94) 19.88 97.19 96.61 97.49
下載: 導出CSV
表 3 五種方法在UCI特征集上的對比
Table 3. Comparison of five methods on the UCI feature set
Model Average accuracy/% Maximum accuracy/% Minimum accuracy/% Modeling time/s SVM 92.77 92.77 92.77 19.41 LSTM 93.35 94.88 92.57 529 SCNs 94.18 94.27 94.06 60.59 L2?SCNs 94.27 94.30 94.10 29.85 NCA?L2?SCNs 97.48 97.93 97.05 18.03
下載: 導出CSV
表 4 NCA?L2?SCNs模型識別結果混淆矩陣
Table 4. Confusion matrix of NCA?L2?SCNs model recognition results
Predicted class Actual class Walking Upstairs Downstairs Sitting Standing Lying Walking 494 23 0 0 0 0 Upstairs 1 443 4 4 0 0 Downstairs 0 2 415 0 0 0 Sitting 1 3 0 466 1 0 Standing 0 0 1 20 531 2 Lying 0 0 0 1 0 535
下載: 導出CSV
www.77susu.com<span id="fpn9h"><noframes id="fpn9h"><span id="fpn9h"></span> <span id="fpn9h"><noframes id="fpn9h"> <th id="fpn9h"></th> <strike id="fpn9h"><noframes id="fpn9h"><strike id="fpn9h"></strike> <th id="fpn9h"><noframes id="fpn9h"> <span id="fpn9h"><video id="fpn9h"></video></span> <ruby id="fpn9h"></ruby> <strike id="fpn9h"><noframes id="fpn9h"><span id="fpn9h"></span> -
參考文獻
[1] Mukherjee D, Mondal R, Singh P K, et al. EnsemConvNet: a deep learning approach for human activity recognition using smartphone sensors for healthcare applications. Multimed Tools Appl, 2020, 79(41-42): 31663 doi: 10.1007/s11042-020-09537-7 [2] Zhuang Z D, Xue Y. Sport-related human activity detection and recognition using a smartwatch. Sensors (Basel) , 2019, 19(22): 5001 doi: 10.3390/s19225001 [3] Ibrahim A A, Küderle A, Ga?ner H, et al. Inertial sensor-based gait parameters reflect patient-reported fatigue in multiple sclerosis. J Neuroeng Rehabilitation, 2020, 17(1): 165 doi: 10.1186/s12984-020-00798-9 [4] Hassan M M, Ullah S, Hossain M S, et al. An end-to-end deep learning model for human activity recognition from highly sparse body sensor data in Internet of Medical Things environment. J Supercomput, 2021, 77(3): 2237 doi: 10.1007/s11227-020-03361-4 [5] Igwe O M, Wang Y, Giakos G C, et al. Human activity recognition in smart environments employing margin setting algorithm. J Ambient Intell Humaniz Comput, 2020: 1 [6] Fang H Q, Tang P, Si H. Feature selections using minimal redundancy maximal relevance algorithm for human activity recognition in smart home environments. J Healthc Eng, 2020, 2020: 1 [7] Heinrich K M, Spencer V, Fehl N, et al. Mission essential fitness: Comparison of functional circuit training to traditional army physical training for active duty military. Mil Med, 2012, 177(10): 1125 doi: 10.7205/MILMED-D-12-00143 [8] Foerster F, Smeja M, Fahrenberg J. Detection of posture and motion by accelerometry: A validation study in ambulatory monitoring. Comput Hum Behav, 1999, 15(5): 571 doi: 10.1016/S0747-5632(99)00037-0 [9] Bharti P, De D, Chellappan S, et al. HuMAn: Complex activity recognition with multi-modal multi-positional body sensing. IEEE Trans Mob Comput, 2018, 18(4): 857 [10] Chen Z H, Jiang C Y, Xie L H. A novel ensemble ELM for human activity recognition using smartphone sensors. IEEE Trans Ind Inform, 2019, 15(5): 2691 doi: 10.1109/TII.2018.2869843 [11] Abidine M B, Fergani B, Menhour I. Activity recognition from smartphones using hybrid classifier PCA-SVM-HMM//2019 International Conference on Wireless Networks and Mobile Communications (WINCOM). Fez, 2019: 1 [12] Mohammad Y, Matsumoto K, Hoashi K. Selecting orientation-insensitive features for activity recognition from accelerometers. IEICE Trans Inf Syst, 2019, E102.D(1): 104 doi: 10.1587/transinf.2018EDP7092 [13] Sansano E, Montoliu R, Belmonte Fernández ó. A study of deep neural networks for human activity recognition. Comput Intell, 2020, 36(3): 1113 doi: 10.1111/coin.12318 [14] Abbaspour S, Fotouhi F, Sedaghatbaf A, et al. A comparative analysis of hybrid deep learning models for human activity recognition. Sensors, 2020, 20(19): 5707 doi: 10.3390/s20195707 [15] Zou Q, Wang Y L, Wang Q, et al. Deep learning-based gait recognition using smartphones in the wild. IEEE Trans Inf Forensics Secur, 2020, 15: 3197 doi: 10.1109/TIFS.2020.2985628 [16] Zhu Q C, Chen Z H, Soh Y C. A novel semisupervised deep learning method for human activity recognition. IEEE Trans Ind Inform, 2019, 15(7): 3821 doi: 10.1109/TII.2018.2889315 [17] Wang D H, Li M. Stochastic configuration networks: Fundamentals and algorithms. IEEE Trans Cybern, 2017, 47(10): 3466 doi: 10.1109/TCYB.2017.2734043 [18] Dai W, Li D P, Chen Q X, et al. Data driven particle size estimation of hematite grinding process using stochastic configuration network with robust technique. J Central South Univ, 2019, 26(1): 43 doi: 10.1007/s11771-019-3981-2 [19] Sheng Z Y, Zeng Z Q, Qu H Q, et al. Optical fiber intrusion signal recognition method based on TSVD-SCN. Opt Fiber Technol, 2019, 48: 270 doi: 10.1016/j.yofte.2019.01.023 [20] Ren Y, Zhang L, Suganthan P N. Ensemble classification and regression-recent developments, applications and future directions. IEEE Comput Intell Mag, 2016, 11(1): 41 doi: 10.1109/MCI.2015.2471235 [21] Zhou H T, Chen J, Dong G M, et al. Bearing fault recognition method based on neighbourhood component analysis and coupled hidden Markov model. Mech Syst Signal Process, 2016, 66-67: 568 doi: 10.1016/j.ymssp.2015.04.037 [22] Qin C, Song S J, Huang G, et al. Unsupervised neighborhood component analysis for clustering. Neurocomputing, 2015, 168: 609 doi: 10.1016/j.neucom.2015.05.064 [23] Zhao L J, Zou S D, Huang M Z, et al. Distributed regularized stochastic configuration networks via the elastic net. Neural Comput Appl, 2021, 33(8): 3281 doi: 10.1007/s00521-020-05178-x [24] Wang W, Wang D H. Prediction of component concentrations in sodium aluminate liquor using stochastic configuration networks. Neural Comput Appl, 2020, 32(17): 13625 doi: 10.1007/s00521-020-04771-4 [25] Song B C, Ra J B. A fast search algorithm for vector quantization using L/sub 2/-norm pyramid of codewords. IEEE Trans Image Process, 2002, 11(1): 10 doi: 10.1109/83.977878 [26] Anguita D, Ghio A, Oneto L, et al. A public domain dataset for human activity recognition using smartphones [DB/OL]. UCI Machine Learning Repository (2013-04)[2021-03-18]. http://archive.ics.uci.edu/ml/datasets/Human+Activity+Recognition+Using+Smartphones -
點擊查看大圖
圖(4) / 表(4)
計量
- 文章訪問數: 584
- HTML全文瀏覽量: 278
- PDF下載量: 76
- 被引次數: 0
