基于人工鱼群算法-极限学习机的多手势精准识别

Multi-gesture accurate recognition based on artificial fish swarm algorithm extreme learning machine

手势作为人机交互的重要方式,因灵活性与便捷性强,已成为控制领域的研究重点。针对上肢康复机器人手势识别技术存在的不足,结合特征组合与滑动窗口法,提出一种基于人工鱼群算法(artificial fish swarm algorithm,AFSA)优化的极限学习机(extreme learning machine,ELM)的多手势精准识别方法,以提高手势识别的准确率。首先,运用表面肌电测量系统采集人体常用的8种手势对应的表面肌电信号(surface electromyography,SEMG),作为后续分类模型的信号源,并运用去噪技术与起止点检测技术对SEMG进行预处理;然后,选取通过主成分分析(principal components analysis,PCA)降维处理后的最优特征组合与最优滑动窗口;接着,采用AFSA搜寻ELM的最优输入权值和隐含阈值,以提高其分类准确率;最后,对AFSA优化的ELM(AFSA-ELM)分类模型、反向传播(back propagation,BP)神经网络分类模型和未优化的ELM分类模型进行比较,以验证所提出方法的精准性。实验结果表明,结合最优特征组合与最优滑动窗口设计的AFSA-ELM分类模型对多种手势的平均识别准确率高达97.4%,比BP神经网络分类模型和未优化的ELM分类模型分别高3.5%和1.6%,验证了所提出方法的识别精准性。研究结果可为手势识别提供新思路,进而为人体上肢动作的深度分析和上肢康复机器人手势识别算法的优化提供理论基础和参考。

As an important way of the human-computer interaction,the gesture has become the research focus in the control field because of its strong flexibility and convenience.Aiming at the shortcomings of gesture recognition technology for the upper limb rehabilitation robot,combined with the feature combination and sliding window method,a multi-gesture accurate recognition method based on the extreme learning machine(ELM)optimized by the artificial fish swarm algorithm(AFSA)was proposed to improve the accuracy of gesture recognition.Firstly,the surface electromyography measurement system was used to collect the surface electromyography(SEMG)corresponding to eight kinds of gestures commonly used by the human as the signal source of the subsequent classification model,and the SEMG was preprocessed by the denoising technology and the start-stop detection technology;then,the optimal feature combination and the optimal sliding window after dimensionality reduction by the principal components analysis(PCA)were selected;then,the AFSA was used to search the optimal input weight and implicit threshold of the ELM to improve its classification accuracy;finally,the ELM classification model optimized by the AFSA(AFSA-ELM),the back propagation(BP)neural network classification model and the non-optimized ELM classification model were compared to verify the accuracy of the proposed method.The experimental results showed that the average recognition accuracy of the AFSA-ELM classification model combined with the optimal feature combination and the optimal sliding window was as high as 97.4%,which was 3.5%and 1.6%higher than that of the BP neural network classification model and the non-optimized ELM classification model,respectively,which verified the recognition accuracy of the proposed method.The research results can provide a new idea for the gesture recognition,and provide a theoretical basis and reference for the depth analysis of human upper limb movement and the optimization of gesture recognition algorithm for upper limb rehabilitation robots.