如何通过猕猴运动皮层的神经元锋电位信号估计其手指移动位置是一神经解码问题,现存方法解决该问题大多采用有监督训练,需要通过训练数据得到神经元锋电位信号与手指移动位置的关系,因此其估计性能依赖于训练数据.本文提出了一种无监督...如何通过猕猴运动皮层的神经元锋电位信号估计其手指移动位置是一神经解码问题,现存方法解决该问题大多采用有监督训练,需要通过训练数据得到神经元锋电位信号与手指移动位置的关系,因此其估计性能依赖于训练数据.本文提出了一种无监督解码方法,该方法基于状态空间模型(State space model,SSM),利用神经网络得到神经元锋电位数与手指移动位置的关系权值,再用逐次状态估计方法去估计手指移动的位置.为减少训练的复杂度和提高估计准确度,采用一种非线性的积分卡尔曼滤波(Cubature Kalman filtering,CKF)来完成神经网络的训练和手指位置的逐次状态估计.与传统方法相比,该方法的最大特点是无监督,可以由神经元锋电位簇向量直接估计手指移动位置,而无需有监督训练.实验结果显示,当采用较少的有监督数据,现存方法与本文方法相比有较大的估计误差;当采用较多的有监督数据,现存方法才具有与本文方法相近似的估计误差.展开更多
Satisfactory results cannot be obtained when three-dimensional (3D) targets with complex maneuvering characteristics are tracked by the commonly used two-dimensional coordinated turn (2DCT) model. To address the probl...Satisfactory results cannot be obtained when three-dimensional (3D) targets with complex maneuvering characteristics are tracked by the commonly used two-dimensional coordinated turn (2DCT) model. To address the problem of 3D target tracking with strong maneuverability, on the basis of the modified three-dimensional variable turn (3DVT) model, an adaptive tracking algorithm is proposed by combining with the cubature Kalman filter (CKF) in this paper. Through ideology of real-time identification, the parameters of the model are changed to adjust the state transition matrix and the state noise covariance matrix. Therefore, states of the target are matched in real-time to achieve the purpose of adaptive tracking. Finally, four simulations are analyzed in different settings by the Monte Carlo method. All results show that the proposed algorithm can update parameters of the model and identify motion characteristics in real-time when targets tracking also has a better tracking accuracy.展开更多
文摘如何通过猕猴运动皮层的神经元锋电位信号估计其手指移动位置是一神经解码问题,现存方法解决该问题大多采用有监督训练,需要通过训练数据得到神经元锋电位信号与手指移动位置的关系,因此其估计性能依赖于训练数据.本文提出了一种无监督解码方法,该方法基于状态空间模型(State space model,SSM),利用神经网络得到神经元锋电位数与手指移动位置的关系权值,再用逐次状态估计方法去估计手指移动的位置.为减少训练的复杂度和提高估计准确度,采用一种非线性的积分卡尔曼滤波(Cubature Kalman filtering,CKF)来完成神经网络的训练和手指位置的逐次状态估计.与传统方法相比,该方法的最大特点是无监督,可以由神经元锋电位簇向量直接估计手指移动位置,而无需有监督训练.实验结果显示,当采用较少的有监督数据,现存方法与本文方法相比有较大的估计误差;当采用较多的有监督数据,现存方法才具有与本文方法相近似的估计误差.
基金supported by the National Natural Science Foundation of China(51467013)
文摘Satisfactory results cannot be obtained when three-dimensional (3D) targets with complex maneuvering characteristics are tracked by the commonly used two-dimensional coordinated turn (2DCT) model. To address the problem of 3D target tracking with strong maneuverability, on the basis of the modified three-dimensional variable turn (3DVT) model, an adaptive tracking algorithm is proposed by combining with the cubature Kalman filter (CKF) in this paper. Through ideology of real-time identification, the parameters of the model are changed to adjust the state transition matrix and the state noise covariance matrix. Therefore, states of the target are matched in real-time to achieve the purpose of adaptive tracking. Finally, four simulations are analyzed in different settings by the Monte Carlo method. All results show that the proposed algorithm can update parameters of the model and identify motion characteristics in real-time when targets tracking also has a better tracking accuracy.
