|
论文摘要目录： |
| 关键词：自动化论文 本科毕业论文 |
摘 要
一些疾病使人们中枢神经系统受到损伤，影响手臂的运动。因此，了解脑皮层控制手臂运动，研究二者之间的关系具有十分重要的意义，将可能使那些渴望运动却无法实现的人们受益非浅。
首先介绍了神经控制与神经元建模的基本思想，描述了如何从生物实验中获得大脑运动皮层的神经元脉冲信号与手臂轨迹坐标数据，并简述了已有的针对这些数据的多种建模方法：群体矢量算法，最优线性滤波，极大似然估计，模式识别和神经网络。然后提出了一种新的建模方法：支持向量机(SVM)回归算法，该算法的核心部分是核函数与两个模型参数的确定。最后，将进行过归一化和神经元集合选取后的实验数据作为样本应用于此模型，并分析经验风险函数。
模型的训练和测试结果表明，上面提出的SVM建模方法是可行的。在给定参数情况下，通过不同核函数情况下的训练结果验证了该模型的可靠性，并用模型的经验风险值和训练过程所用的的支持矢量个数对模型进行评估与比较。比较结果显示，线性回归结果与RBF核(p1=10)非线性回归结果精度较高；选取部分神经元脉冲集合建模时比全部神经元脉冲集合参与建模时的经验风险小一些； 值的变化会影响模型的训练结果、风险误差值和回归所用的支持向量个数等，且随着 值的增大，回归所用的支持向量个数逐渐减少。不同训练样本集情况下的预测结果表明，由不同训练样本得到的模型的推广性能具有较大差异。
关键词：脑皮层控制，神经元脉冲，支持向量机，回归，经验风险。
Abstract
Some people can not move their arms because diseases or injuries affect the central nervous systems of them. So better knowledge of the extraction algorithms for cortical control of arm prosthetics can help people who want to move but have no ability.
In this paper, we introduce the concept of neural control and neurons modeling, and describe the way to get firing rates of motor cortex and accurate arm trajectories from biology experiments. We present some algorithms for modeling, such as population vector algorithm (PV), optimal linear filters, maximum likelihood estimation, pattern recognition and neural networks. Then we propose a new modeling method: Support vector machine (SVM), especially the support vector regression, kernel function. Kernel function and two parameters are most important for the algorithm. We deal with the data from experiments and sel无忧论文 【http://www.uklunwen.com】ect parts of neurons sets, and analyze the empirical risk.
According to the results of training and testing of the model, SVM regression presented above is a reliable algorithm here. The dependency of the model is demonstrated using the training results with different kernel functions. We estimate the model from the empirical risk and the number of support vectors in SVM training. According the result of comparisons, the training results of linear regression and nonlinear regression of RBF kernel with the width of 10 are better. The empirical risk of model with parts of neurons is less than that with all the neurons. If the value of changes, the training result, empirical risk and number of support vectors used in regression will not invariable. As the increasing of value, the number of support vectors used in regression will be fewer. The results of testing illustrate that the generalization performances of the models with distinct training data sets are much different.
Keywords：cortical control, firing rates, support vector machine, regression, empirical risk
|
|
|
第1页 第2页 |
|
|
|
| 上一篇：基于ARM-LINUX的无线终端应用下一篇：基于混沌的数字图像加密
|
| 最新论文 |
最热门论文 |
|
|
|
|
|
