表面肌电信号作为一种记录肌肉收缩活动的生物电信号，为研究和分析人体肌肉功能提供了一种非侵入式的手段。随着人工智能的发展，基于肌电信号的手势识别被广泛应用于工业自动化、康复医学以及虚拟现实等领域，其中工业自动化包含控制机器人、远程控制生产设备，康复医学包含假肢控制、运动辅助，虚拟现实包含手势导航或控制、手势交互游戏。在手势识别的研究过程中，采集数据、预处理、特征提取和分类是实现高效、准确识别的关键步骤。本文主要在特征提取、分类识别进行了深入研究，并提出了改进方法。主要的研究工作如下：

     （1）在机械手臂的手势识别中，针对特征提取方法提取到的特征维度高的问题，本文提出了基于相关性热力图的特征降维方法，旨在通过降低特征维度的同时提取到数据的核心信息，以提高后续分类任务的效率和准确性。该方法的核心思想是利用统计学中的协方差和方差来量化sEMG中不同通道之间的相关性，从而筛选出具有代表性的特征。为了验证所提方法的有效性，在特征降维后，本文采用了三种传统的机器学习分类算法对降维后的特征进行了分类处理，并对这三种特征降维方法的性能进行了全面评估。实验结果表明，通过相关性热力图降维后，使用随机森林分类的准确率为81.6%，证明了所提出的降维方法在有效减少特征数量的同时，能够显著提升表面肌电信号分类的准确性。

     （2）针对手势识别算法中模型分类准确率低、网络结构冗余的问题，本文提出了基于双卷积神经网络的分类算法。首先，通过小波阈值去噪对原始肌电信号进行预处理，这是因为肌电信号采集过程中会受到各种噪声的影响，这些噪声会严重影响信号的质量和后续处理的准确性；其次，为了能够提取到更全面、更抽象的特征信息，将三层双卷积层和最大池化层结合，从输入的原始信号中精确提取、优化过滤出有明显差别的特征图；最后，在特征提取和优化过程完成之后，通过引入全连接层的设计，将之前双卷积层和最大池化层处理后的高维特征映射转化为一维向量形式。全连接层的加入，进一步提高了网络处理不同信号分类问题时的适应性和效率，确保整体模型在高维特征空间中的映射能力和训练效率。实验结果表明，DUAL-CNN在单通道上的平均准确率为95.91%，与使用同一数据集的其他算法相比，该算法提高了整体手势分类的准确率。

       本文所提出的特征降维和分类识别算法，有效提升了基于表面肌电信号的手势动作识别的性能，这不仅为残疾人士提供了新的希望，还展示了科技进步给社会带来的积极影响，推动了人机交互技术的发展。未来，基于表面肌电信号的手势识别技术将在更多领域发挥更大的作用。

    Surface electromyography, as a kind of bioelectrical signals that record muscle contraction activities, provide a non-invasive means to study and analyze human muscle functions. With the development of artificial intelligence, gesture recognition based on EMG signals is widely used in the fields of industrial automation, rehabilitation medicine, and virtual reality, where industrial automation includes controlling robots and remotely controlling production equipment, rehabilitation medicine includes prosthetic limb control and motion assistance, and virtual reality includes gesture navigation or control, and gesture-interactive games. In the research process of gesture recognition, data acquisition, preprocessing, feature extraction and classification are the key steps to realize efficient and accurate recognition. In this paper, two methods are improved in feature extraction and classification recognition to improve the accuracy and efficiency of gesture recognition, and the main research work is as follows:

    (1) In gesture recognition of robotic arms, for the problem of high dimensionality of features extracted by feature extraction methods, this paper proposes a feature dimensionality reduction method based on correlation heat map, which aims to extract the core information of the data by reducing the dimensionality of the features at the same time, in order to improve the efficiency and accuracy of the subsequent classification tasks. The core idea of the method is to utilize the covariance and variance in statistics to quantify the correlation between different channels in the sEMG, so as to filter out representative features. In order to verify the effectiveness of the proposed method, after feature dimensionality reduction, the paper employs three traditional machine learning classification algorithms to categorize the dimensionality reduced features and comprehensively evaluates the performance of these three feature dimensionality reduction methods. The experimental results show that the accuracy of classification using random forest is 81.6% after dimensionality reduction by correlation heat map, which proves that the proposed dimensionality reduction method can significantly improve the accuracy of sEMG signal classification while effectively reducing the number of features.

    (2)Aiming at the problems of low model classification accuracy and redundant network structure in gesture recognition algorithms, the paper proposes a classification algorithm based on dual convolutional neural networks. Firstly, the original EMG signal is preprocessed by wavelet threshold denoising, because the EMG signal acquisition process will be affected by a variety of noises, which will seriously affect the quality of the signal and the accuracy of the subsequent processing. Secondly, in order to be able to extract a more comprehensive and higher abstract feature information, the three-layer dual convolutional layer and the maximal pooling layer are combined, and the feature maps are accurately extracted, optimized and filtered out from the input original signal. Finally, after the feature extraction and optimization process is completed, the feature maps with significant differences are extracted and optimized by the dual convolutional neural networks, filter out the feature maps with obvious differences. After the feature extraction and optimization process is completed, the high-dimensional feature maps previously processed by the dual convolutional layer and the maximum pooling layer are transformed into a one-dimensional vector form by the introduction of the fully connected layer design. The addition of the fully connected layer further improves the adaptability and efficiency of the network when dealing with different signal classification problems, and ensures the mapping ability and training efficiency of the overall model in the high-dimensional feature space. The experimental results show that DUAL-CNN has an average accuracy of 95.91% on a single channel, which improves the overall gesture classification accuracy compared to other algorithms using the same dataset.

   The feature reduction and classification recognition algorithm proposed in the paper effectively improves the performance of gesture action recognition based on sEMG signal, which not only provides a new hope for people with disabilities, but also promotes the development of human-computer interaction technology and demonstrates the positive impact of scientific and technological progress on society. In the future, sEMG signal-based gesture recognition technology will play a greater role in more fields.

[1]苏芳茵. 表面肌电信号分析及运动模式分类研究[D]. 厦门: 华侨大学, 2016.

[2]张志勇. 肌电信号采集与肌电假肢控制的研究[D]. 哈尔滨: 哈尔滨工业大学, 2010.

[3]Alsayegh O A. EMG-based human-machine interface system[C]//IEEE International Conference on Multimedia and Expo. Proceedings. Latest Advances in the Fast Changing World of Multimedia. IEEE, 2000, 2: 925-928.

[4]Park S H, Lee S P. EMG pattern recognition based on artificial intelligence techniques[J]. IEEE Transactions on Rehabilitation Engineering, 1998, 6(4): 400-405.

[5]李阳. 智能仿生手臂肌电信号—运动模型化与模式识别理论方法研究[D]. 长春: 吉林大学, 2012.

[6]丁其川, 熊安斌, 赵新刚, 等. 基于表面肌电的运动意图识别方法研究及应用综述[J].自动化学报, 2016, 42(1): 13-25.

[7]周杨. 基于深度学习的肌电手势识别算法研究[D]. 南京: 南京邮电大学, 2023.

[8]Khezri M, Jahed M. An exploratory study to design a novel hand movement identification system[J]. Computers in Biology and Medicine, 2009, 39(5): 433-442.

[9]Jiang N, Dosen S, Muller K R, et al. Myoelectric control of artificial limbs-is there a need to change focus?[In the spotlight][J]. IEEE Signal Processing Magazine, 2012, 29(5): 152-150.

[10]Wang W, Stefano A D, Allen R. A simulation model of the surface EMG signal for analysis of muscle activity during the gait cycle[J]. Computers in Biology & Medicine, 2006, 36(6): 601-618.

[11]Ghazwan A, Forrest S M, Holt C A, et al. Can activities of daily living contribute to EMG normalization for gait analysis?[J]. PLoS ONE, 2017, 12(4): e0174670.

[12]Norcross M F, Blackburn J T, Goerger B M. Reliability and interpretation of single leg stance and maximum voluntary isometric contraction methods of electromyography normalization[J]. Journal of electromyography and kinesiology, 2010, 20(3): 420-425.

[13]Burden A, Bartlett R. Normalisation of EMG amplitude: an evaluation and comparison of old and new methods[J]. Medical Engineering & Physics, 1999, 21(4): 247-257.

[14]Benoit D L, Lamontagne M, CerulliG et al. The clinical significance of electromyography normalisation techniques in subjects with anterior cruciate ligament injury during treadmill walking[J]. Gait & Posture, 2003, 18(2): 56-63.

[15]Papagiannis G I, Triantafyllou A I, Roumpelakis I M, et al. Methodology of surface electromyography in gait analysis: review of the literature[J]. Journal of medical engineering & technology, 2019, 43(1): 59-65.

[16]Al-Faiz M Z, Ali A A, Miry A H. A k-nearest neighbor based algorithm for human arm movements recognition using EMG signals[C]//1st International Conference on Energy, Power and Control (EPC-IQ). IEEE, 2010: 159-167.

[17]Rasheed S, Stashuk D, Kamel M. Adaptive fuzzy k-NN classifier for EMG signal decomposition[J]. Medical engineering & physics, 2006, 28(7): 694-709.

[18]Camargo J, Young A. Feature selection and non-linear classifiers: effects on simultaneous motion recognition in upper limb[J]. IEEE Transactions on Neural Systems and Rehabilitation on Engineering, 2019, 27(4): 743-750.

[19]Guyon I, Weston J, Barnhill S, et al. Gene Selection for Cancer Classification using Support Vector Machines[J]. Machine Learning, 2002, 46(1-3): 389-422.

[20]周志华. 机器学习[M]. 北京: 清华大学出版社, 2016: 133-140.

[21]Saikia A, Mazumdar S, Sahai N, et al. Performance analysis of artifcial neural network for hand movement detection from EMG signals[J]. IETE Journal Research, 2022, 68(2): 1074-1083.

[22]Shahzaib M, Shakil S, Ghufar S, et al. Classifcation of forearm EMG signals for 10 motions using optimum feature-channel combinations[J]. Computer Methods in Biomechanics and Biomedical Engineering, 2021, 24(9): 945-955.

[23]Nazmi N, Abdul Rahman M A, Yamamoto S I, et al. A review of classification techniques of EMG signals during isotonic and isometric contractions[J]. Sensors, 2016, 16(8): 1304.

[24]Kendell C, Lemaire ED, Losier Y, et al. A novel approach to surface electromyography: an exploratory study of electrode-pair selection based on signal characteristics[J]. Journal of neuroengineering and rehabilitation, 2012, 9: 1-8.

[25]Krasoulis A, Vijayakumar S, Nazarpour K. Multi-grip classification-based prosthesis control with two EMG-IMU sensors[J]. IEEE Transactions on Neural Systems and Rehabilitation on Engineering, 2020, 28(2): 508-518.

[26]Adewuyi A A, Hargrove L J, Kuiken T A. An analysis of intrinsic and extrinsic hand muscle EMG for improved pattern recognition control[J]. IEEE Transactions on Neural Systems and Rehabilitation on Engineering, 2015, 24(4): 485-494.

[27]仝润泽. 面向手势识别的表面肌电信号分类算法研究[D]. 浙江: 浙江工业大学, 2020.

[28]Khushaba R N, Kodagoda S. Electromyogram (EMG) feature reduction using mutual components analysis for multifunction prosthetic fingers control[C]//12th International Conference on Control Automation Robotics & Vision (ICARCV). IEEE, 2012: 1534-1539.

[29]Nishad A, Upadhyay A, Pachori R B, et al. Automated classification of hand movements using tunable-Q wavelet transform based filter-bank with surface electromyogram signals[J]. Future Generation Computer Systems, 2019, 93: 96-110.

[30]Bird J J, Pritchard M G, Fratini A, et al. Synthetic Biological Signals Machine-Generated by GPT-2 Improve the Classification of EEG and EMG Through Data Augmentation[J]. IEEE Robotics and Automation Letters, 2021, 6(2): 3498-3504.

[31]Leone F, Gentile C, Ciancio A L, et al. Simultaneous sEMG Classification of Hand/Wrist Gestures and Forces[J]. Frontiers in Neurorobotics, 2019, 13: 42.

[32]Copaci D, Arias J, Gómez-Tomé M, et al. sEMG-based gesture classifier for a rehabilitation glove[J]. Frontiers in Neurorobotics, 2022, 16: 750482.

[33]Stefanou T, Guiraud D, Fattal C, et al. Frequency-domain sEMG classification using a single sensor[J]. Sensors, 2022, 22(5): 1939.

[34]Kim M, Chung W K, Kim K. Preliminary study of virtual semg signal-assisted classification[C]//2019 IEEE 16th International Conference on Rehabilitation Robotics (ICORR). IEEE, 2019: 1133-1138.

[35]Ma F, Song F, Liu Y , et al. sEMG-Based Neural Network Prediction Model Selection of Gesture Fatigue and Dataset Optimization[J]. Computational Intelligence and Neuroscience, 2020, 2020(1): 1-17.

[36]Fatayer A, Gao W, Fu Y. sEMG-Based Gesture Recognition Using Deep Learning From Noisy Labels[J]. IEEE Journal Biomedical Health Information, 2022, 26(9): 4462-4473.

[37]Manfredo A, Matteo C, Henning M. Deep Learning with Convolutional Neural Networks Applied to Electromyography Data: A Resource for the Classification of Movements for Prosthetic Hands[J]. Frontiers in neurorobotics, 2016, 10: 9.

[38]Geng W, Du Y, Jin W, et al. Gesture recognition by instantaneous surface EMG images[J]. Scientific reports, 2016, 6: 36571.

[39]Xia P, Hu J, Peng Y. EMG-based estimation of limb movement using deep learning with recurrent convolutional neural networks[J]. Artificial organs, 2018, 42(5): E67-E77.

[40]Atzori M, Cognolato M, Müller H. Deep learning with convolutional neural networks applied to electromyography data: A resource for the classification of movements for prosthetic hands[J]. Frontiers in neurorobotics, 2016, 10: 9.

[41]Wentao Wei, Yongkang Wong, Yu Du, et al. A multi-stream convolutional neural network for sEMG-based gesture recognition in muscle-computer interface[J]. Pattern Recognition Letters, 2019(119): 131-138.

[42]Zia ur Rehman M, Waris A, Gilani S O, et al. Multiday EMG-based classification of hand motions with deep learning techniques[J]. Sensors, 2018, 18(8): 2497.

[43]Li X, Samuel O W, Zhang X, et al. A motion-classification strategy based on

sEMG-EEG signal combination for upper-limb amputees[J]. Journal of Neuroengineering & Rehabilitation, 2017, 14(1): 2

[44]杨彬, 黄鹏程, 杨庆华, 等. 基于肌电控制的灵巧手及其控制系统设计[J]. 机电工程, 2016, 33(4): 388-394.

[45]尤波, 刘意, 赵汗青, 等. 人体肌电信号的生理层模型[J]. 计算机仿真, 2015, 32(1): 367-371.

[46]Sun Y, L C, Li G, et al. Gesture recognition based on kinect and s EMG signal fusion[J]. Mobile Networks and Applications, 2018, 23: 797-805.

[47]Meltzner G S, Heaton J T, Deng Y, et al. Development of s EMG sensors and algorithms for silent speech recognition[J]. Journal of Neural Engineering, 2018, 15(4): 046031.

[48]朱昊, 辛长宇, 吉小军, 等. 表面肌电信号前端处理电路与采集系统设计[J]. 测控技术, 2008, 27(3): 37-39.

[49]侯文生, 万莎, 吴小鹰, 等. 手指运动姿态检测及假肢手指动作实时控制[J]. 重庆大学学报, 2013, 36(4): 103-109.

[50]王聪. 基于RGB-D相机的AGV三维环境感知重建和路径规划研究[D]. 南昌: 南昌大学, 2022.

[51]Chu J U, Moon I, Lee Y J, et al. A Supervised Feature-Projection-Based Real-Time EMG Pattern Recognition for Multifunction Myoelectric Hand Control[J]. IEEE/ASME Transactions on Mechatronics, 2007, 12(3): 282-290.

[52]钱晓进, 郑如冰, 王传林. 高性能肌电检测前置放大器的设计[J]. 现代科学仪器, 2003, (3): 50-52.

[53]陈斌. 基于表面肌电信号的手势识别研究[D]. 南昌: 南昌大学, 2021.

[54]韦晔. 基于Myo的手势和手臂运动识别研究[D]. 沈阳: 沈阳工业大学, 2021.

[55]Zhang X, Zhang M. Study on the methods of feature extraction based on electromyographic signal classification[J]. Medical & Biological Engineering & Computing, 2023, 61(7): 1773-1781.

[56]Li K X, Zhang J H, Wang L F, et al. A review of the key technologies for s EMG-based human-robot interaction systems[J]. Biomedical Signal Processing and Control, 2020, 62: 1-17.

[57]Jamal M Z. Signal acquisition using surface EMG and circuit design considerations for robotic prosthesis[J]. Computational Intelligence in Electromyography Analysis-A Perspective on Current Applications and Future Challenges, 2012, 18: 427-448.

[58]张廷尉, 陈红, 王磊. 基于 Matlab的巴特沃斯数字低通滤波器的设计[J]. 鞍山师范学院学报, 2012, 14(2): 13-15.

[59]卢朝林. 表面肌电信号采集臂带的设计与手势识别应用[D]. 成都: 电子科技大学, 2020.

[60]Li L, Wang J H, Ban S. A design method for IIR and FIR digital notch filter used to s EMG[J]. Applied Mechanics & Materials, 2013, 263: 184-187.

[61]于然, 付周兴, 王清亮. 基于MATLAB的电弧建模仿真及故障分析[J]. 高压电器, 2011, (9): 95-99.

[62]李航. 统计学习方法[M]. 北京: 清华大学出版社, 2012: 95-135.

[63]Saikia A, Mazumdar S, Sahai N et al. Performance analysis of artifcial neural network for hand movement detection from EMG signals[J]. IETE Journal of Research, 2022, 68(2):1074-1083.

[64]黄晓娟, 张莉. 改进的多类支持向量机递归特征消除在癌症多分类中的应用[J]. 计算机应用, 2015, 35(10): 2798-2802.

[65]Elbeshbeshy A M, Rushdi M A, El-Metwally S M. Electromyography signal analysis and classification using time-frequency representations and deep learning[C]//43rd Annual International Conference of the IEEE Engineering in Medicine & Biology Society (EMBC). IEEE, 2021: 661-664.

[66]Cover T, Hart P. Nearest neighbor pattern classification[J]. IEEE transactions on information theory, 1967, 13(1): 21-27.

[67]唐胡鑫. 网站运营与数据分析双色版[M]. 北京: 航空工业出版社, 2016: 81-82.

[68]黄勤龙. 云计算数据安全[M], 北京: 北京邮电大学出版社, 2018: 95.

[69]Vapnik V N, Lerner A Y. Recognition of patterns with help of generalized portraits[J]. Avtomat. i Telemekh, 1963, 24(6): 774-780.

[70]Vapnik,V. Statistical learning theory[J]. Wiley, 1998, 3: 401-492.

[71]边肇祺. 模式识别[M]. 清华大学出版社有限公司, 2000.

[72]Hsieh W W. Machine learning methods in the environmental sciences: Neural networks and kernels[M]. Cambridge university press, 2009.

[73]Breiman L. Random forests[J]. Machine learning, 2001, 45: 5-32.

[74]Lobov S, Krilova N, Kastalskiy I, et al. Latent factors limiting the performance of sEMG-interfaces[J]. Sensors, 2018, 18(4): 1122.

[75]Sapsanis C. Recognition of basic hand movements using electromyography[D]. University of Patras (Greece), 2013.

[76]Sapsanis C, Georgoulas G, Tzes A, et al. Improving EMG based classification of basic hand movements using EMD[C]//35th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC). IEEE, 2013: 5754-5757.

[77]Allard U C, Nougarou F, Fall C L, et al. A convolutional neural network for robotic arm guidance using sEMG based frequency-features[C]//IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). IEEE, 2016: 2464-2470.

[78]窦慧, 张凌茗, 韩峰, 等. 卷积神经网络的可解释性研究综述[J]. 软件学报, 2023, 35(1): 159-184.

[79]Fasel B. Robust face analysis using convolutional neural networks[C]//International Conference on Pattern Recognition. IEEE, 2002, 2: 40-43.

[80]黄妤宁. 非合作通信环境下飞机声信号类型识别的方法研究[D]. 哈尔滨: 哈尔滨工程大学, 2019.

[81]Apicella A, Donnarumma F, Isgrò F, et al. A survey on modern trainable activation functions[J]. Neural Networks, 2021, 138: 14-32.

[82]陈永生, 喻玲娟, 谢晓春. 基于全卷积神经网络的 SAR 图像目标分类[J]. 雷达科学与技术, 2018, 16(3): 242-248.

[83]Iqbal O, Fattah S A, Zahin S. Hand movement recognition based on singular value decomposition of surface EMG signal[C]//IEEE Region 10 Humanitarian Technology Conference (R10-HTC). IEEE, 2017: 837-842.

[84]Zhi-Hua, Zhou, Ji, et al. Deep forest[J]. National Science Review, 2019, 6(1): 74-86.

[85]Toledo-Perez D C, Rodriguez-Resendiz J, Gomez-Loenzo R A. A study of computing zero crossing methods and an improved proposal for EMG signals[J]. IEEE Access, 2020, 8: 8783-8790.

[86]Cokun M, Yldrm Z. Efficient deep neural network model for classification of grasp types using sEMG signals[J]. Journal Ambient Intelligence and Humanized Computing, 2021(7): 4437-4450.