CN103164026A

CN103164026A - Method and device of brain-computer interface based on characteristics of box dimension and fractal intercept

Info

Publication number: CN103164026A
Application number: CN2013100957780A
Authority: CN
Inventors: 周卫东; 孙中钱; 田帅; 王芸; 袁琦; 高诺
Original assignee: Shandong University
Current assignee: Shandong University
Priority date: 2013-03-22
Filing date: 2013-03-22
Publication date: 2013-06-19
Anticipated expiration: 2033-03-22
Also published as: CN103164026B

Abstract

The invention relates to a method and a device of a brain-computer interface based on characteristics of box dimension and fractal intercept. Detection to electroencephalogram states can be achieved on the basis of a hardware platform which is formed by an electroencephalogram amplifier and a computer. Firstly, electroencephalogram signals are collected by the electroencephalogram amplifier and a data collection card, and then the collected electroencephalogram signals are sent to the computer so as to be processed, so that extraction of the characteristics of the box dimension and the fractal intercept is achieved, and classification to the electroencephalogram signals is finished through a Boosting classifier. According to the method and the device, extraction of the characteristics of the electroencephalogram signals is achieved by utilizing the box dimension and the fractal intercept with good characteristic effects, and the electroencephalogram signals are classified by the Boosting classifier, and therefore classified marks for electroencephalogram signals of brain-computer interface motor imagery are obtained.

Description

Brain-computer interface method and device based on box dimension and fractal intercept features

技术领域technical field

本发明涉及基于盒维和分形截距特征的脑机接口方法及装置，属于脑机接口的技术领域。The invention relates to a brain-computer interface method and device based on box dimension and fractal intercept features, and belongs to the technical field of brain-computer interface.

技术背景technical background

现实生活中有很多病人因患有严重的运动障碍，比如像脊髓损伤或肌肉萎缩性脊髓侧索硬化症(ALS)等，而丧失基本的与外界进行语言或者肢体沟通的能力。这严重影响了患者的生活质量，也给其家庭和社会造成重大的负担。脑机接口(BCI)是通过人脑和外界之间建立不依赖于常规大脑信息输出通路的一种人机交互系统。脑机接口技术在康复医疗、军事等诸多领域都有着广泛的应用。In real life, many patients suffer from severe movement disorders, such as spinal cord injury or amyotrophic lateral sclerosis (ALS), and lose the basic ability to communicate with the outside world through language or body. This has seriously affected the quality of life of patients, and also caused a significant burden to their families and society. Brain-computer interface (BCI) is a human-computer interaction system that establishes between the human brain and the outside world without relying on the conventional brain information output pathway. Brain-computer interface technology has a wide range of applications in rehabilitation medicine, military and many other fields.

不同的肢体部位运动所激活的大脑皮层区域也各不相同；单边肢体运动或想象运动能激活主要的感觉运动皮层，大脑对侧产生事件相关去同步电位ERD(Event RelatedDesynchronization)，大脑同侧产生事件相关同步电位ERS(Event Related Synchronization)；ERD是指当某一皮层区域活跃起来时，特定频率的节律性活动表现出幅度的降低，ERS是指当某一活动在一定时刻没有使相关皮层区域明显地活跃起来，特定频率就表现出幅度升高。电生理学研究表明，运动想象会导致脑电节律的变化。运动想象会导致频率为8-12Hz的u节律和频率为13-28Hz的β节律的幅度压制即事件相关去同步化ERD，或幅度增加即事件相关同步ERS。The cerebral cortex regions activated by different limb movements are also different; unilateral limb movement or imaginary movement can activate the main sensorimotor cortex, and the opposite side of the brain generates event-related desynchronization potential ERD (Event Related Desynchronization), and the same side of the brain generates Event-related synchronous potential ERS (Event Related Synchronization); ERD means that when a certain cortical area is active, the rhythmic activity of a specific frequency shows a decrease in amplitude, and ERS means that when a certain activity does not make the relevant cortical area active at a certain moment. Significantly active, specific frequencies exhibit increased amplitude. Electrophysiological studies have shown that motor imagery causes changes in brain electrical rhythms. Motor imagery can lead to amplitude suppression of the u-rhythm with a frequency of 8-12 Hz and a beta rhythm with a frequency of 13-28 Hz (event-related desynchronization ERD), or amplitude increase (event-related synchronous ERS).

BCI技术通过提取使用者的脑电信息，然后利用一些机器算法将大脑的不同状态转化为控制性命令，进而实现对外部设备的控制。BCI的目的是建立一个能够帮助用户直接与外界进行交流的系统，而不用借助于传统的神经肌肉途径，其中，寻求有效的特征提取方法是提高识别率的关键技术之一。相同的特征使用不同的分类器进行分类，所得的结果也会有所不同。因此，在选择特征的同时，分类器的选择也至关重要。BCI technology extracts the user's EEG information, and then uses some machine algorithms to convert different states of the brain into control commands, and then realizes the control of external devices. The purpose of BCI is to build a system that can help users communicate directly with the outside world without resorting to traditional neuromuscular pathways. Among them, seeking an effective feature extraction method is one of the key technologies to improve the recognition rate. The same feature is classified by different classifiers, and the results obtained will be different. Therefore, while selecting features, the choice of classifier is also crucial.

目前已有多种特征提取的方法，如自适应的共空域模式、频带功率、AR模型等。2007年，Hammon PS等人在IEEE Transactions on Biomedical Engineering上发表的论文“Preprocessing and meta-classification for brain-computer interfaces”提出一种预处理和多分类器的方法，取得了较好的结果。但是，该方法的预处理和后处理都比较复杂，增加了该方法实现的难度，另一方面也很大程度上降低了方法执行的速度。At present, there are many methods of feature extraction, such as adaptive common space mode, frequency band power, AR model, etc. In 2007, the paper "Preprocessing and meta-classification for brain-computer interfaces" published by Hammon PS et al. on IEEE Transactions on Biomedical Engineering proposed a preprocessing and multi-classifier method, which achieved good results. However, the pre-processing and post-processing of this method are relatively complicated, which increases the difficulty of implementing the method, and on the other hand reduces the execution speed of the method to a large extent.

发明内容Contents of the invention

针对现有技术的不足，本发明提出一种基于盒维和分形截距特征的脑机接口方法。该方法是将提取到的脑电信号盒维和分形截距特征作为输入参数，送入Boosting分类器中进行分类，进而获得脑电状态检测结果。Aiming at the deficiencies of the prior art, the present invention proposes a brain-computer interface method based on box dimension and fractal intercept features. In this method, the extracted EEG signal box dimension and fractal intercept feature are used as input parameters, and then sent to the Boosting classifier for classification, and then the EEG state detection result is obtained.

本发明还提供一种执行上述基于盒维和分形截距特征的脑机接口方法的装置。The present invention also provides a device for implementing the brain-computer interface method based on the box dimension and fractal intercept features.

发明概述：Summary of the invention:

一种基于盒维和分形截距特征的脑机接口方法是基于脑电放大器和计算机构成的硬件平台实现对脑电状态的检测；首先通过脑电放大器和数据采集卡采集脑电信号，然后将采集到的脑电信号送至计算机进行处理，实现盒维和分形截距的特征提取，并通过Boosting分类器完成对脑电信号的分类,发出控制命令。A brain-computer interface method based on box dimension and fractal intercept features is based on a hardware platform composed of an EEG amplifier and a computer to detect the EEG state; firstly, the EEG signal is collected through the EEG amplifier and the data acquisition card, and then The received EEG signals are sent to the computer for processing to realize the feature extraction of box dimension and fractal intercept, and the Boosting classifier is used to complete the classification of EEG signals and issue control commands.

发明详述：Detailed description of the invention:

本发明的技术方案如下：Technical scheme of the present invention is as follows:

一种基于盒维和分形截距特征的脑机接口方法，包括以下步骤：A brain-computer interface method based on box dimension and fractal intercept features, comprising the following steps:

1）采集实验者想象左小指、舌头运动时，大脑所产生的脑电信号，采样频率为1000Hz；实验者想象左小指运动时其脑电信号对应的类别标识为0类，想象舌头运动时其脑电信号对应的类别标识为1类；1) Collect the EEG signals generated by the brain when the experimenter imagines the movement of the left little finger and tongue, and the sampling frequency is 1000 Hz; The category corresponding to the EEG signal is identified as category 1;

2）对采集到的脑电信号进行降采样，采样频率为100Hz；2) Down-sampling the collected EEG signal, the sampling frequency is 100Hz;

3）对经步骤2）降采样后的脑电信号进行8-30Hz的带通滤波；3) Perform 8-30 Hz band-pass filtering on the EEG signal after step 2) downsampling;

4）提取脑电信号中各通道的盒维和分形截距特征，其中对所述的提取脑电信号各通道盒维和分形截距的步骤方法为：4) Extract the box dimension and fractal intercept features of each channel in the EEG signal, wherein the steps for extracting the box dimension and fractal intercept of each channel of the EEG signal are:

a)将降采样和滤波后长度为L点的脑电信号S，平均分割成长为H点的G个子段，然后计算每一个子段的盒维和分形截距；a) After downsampling and filtering, the EEG signal S with a length of L points is averagely divided into G sub-segments of H points, and then the box dimension and fractal intercept of each sub-segment are calculated;

b)对脑电信号S的每个子段，将其继续分割成长为r的T个小段（r＝2^h,(0＜h＜log₂H)，T等于H/r的整数部分），取覆盖信号的盒子的边长等于r。对第i个小段（其中i＝1,…，T），若其振幅的最小值和最大值分别落在第k个和第l个盒子中，则覆盖第i个小段所需的盒子数n(i)为：b) For each sub-segment of the EEG signal S, continue to divide it into T small segments of r (r=2 ^h , (0<h<log ₂ H), T is equal to the integer part of H/r), take The side length of the box covering the signal is equal to r. For the i-th small segment (where i=1,...,T), if the minimum and maximum amplitudes fall in the k-th and l-th boxes respectively, then the number of boxes n required to cover the i-th small segment (i) is:

n(i)＝l-k+1；n(i)=l-k+1;

c)覆盖该子段所需的盒子总数Num(r)为：c) The total number of boxes Num(r) required to cover this subsection is:

$Num Num ((r r)) = = {Σ Σ}_{i i = = 00}^{T T} n no ((i i))$

d)该子段脑电信号的盒维数D为：d) The box dimension D of the sub-segment EEG signal is:

$D D. = = \underset{r r &RightArrow; &Right Arrow; 00}{lim lim} \frac{log log Num Num ((r r))}{log log ((11 / / r r))}$

e)当盒子的边长r变化时，步骤d)中所述的公式满足直线方程：e) When the side length r of the box changes, the formula described in step d) satisfies the equation of a straight line:

logNum(r)＝D·log(1r)+ClogNum(r)=D·log(1r)+C

其中，直线的斜率为D，截距为C；取不同的r值，计算得到若干组(r,Num(r))；应用最小二乘曲线拟合算法，求得该直线方程的斜率D和截距C；这里，斜率D即为该子段脑电信号的盒维数，而截距C则为该子段脑电信号的分形截距；Among them, the slope of the straight line is D, and the intercept is C; different r values are taken to calculate several groups (r, Num(r)); the slope D and the slope of the straight line equation are obtained by applying the least squares curve fitting algorithm Intercept C; Here, the slope D is the box dimension of the sub-segment EEG signal, and the intercept C is the fractal intercept of the sub-segment EEG signal;

5）将步骤4）提取到的盒维和分形截距特征输入到Boosting分类器进行分类，得到输出概率值；5) Input the box dimension and fractal intercept feature extracted in step 4) to the Boosting classifier for classification, and obtain the output probability value;

6）将输出概率值与预设阈值进行比较，其中所述的预设阈值为0.5，获得脑电状态检测结果并转换为控制命令：6) Comparing the output probability value with the preset threshold value, wherein the preset threshold value is 0.5, obtaining the EEG state detection result and converting it into a control command:

当输出概率值大于预设阈值时，则判断此时的脑电状态为想象舌头时的脑电信号，并转换为控制命令1；When the output probability value is greater than the preset threshold, it is judged that the EEG state at this time is the EEG signal when the tongue is imagined, and converted into a control command 1;

当输出概率值小于或等于预设阈值时，则判断此时的脑电状态为想象左小指时的脑电信号，并转换为控制命令2；When the output probability value is less than or equal to the preset threshold, it is judged that the EEG state at this time is the EEG signal when imagining the left little finger, and it is converted into a control command 2;

步骤3）中所述的对脑电信号进行滤波的方法，包括步骤如下：The method for filtering the EEG signal described in step 3) includes the following steps:

对脑电信号利用J阶的巴特沃斯滤波器进行8-30Hz的带通滤波，优选J=5；Use a J-order Butterworth filter to perform band-pass filtering of 8-30 Hz on the EEG signal, preferably J=5;

步骤5）中所述的Boosting分类器的实现步骤为：The implementation steps of the Boosting classifier described in step 5) are:

a)分类器训练所使用的训练数据特征集X={X_j∈R^K,j=1,…,N}，其对应的标识Y={y_j∈{0,1},j=1,…,N}，其中，K=Ch×s是特征的数目，其中Ch是通道数，而s是单次实验单个导联上所提取的特征向量的数目，N为训练数据中所包含的单次实验的数目；F_m表示m步后的分类器；设定迭代次数为M；设定第j次单次实验的脑电信号特征向量X_j为想象舌头的初始概率P₀(y_j＝1|X_j)＝0.5,j＝1,…,N，设定第j次单次实验的脑电信号特征向量X_j的初始分类器为F₀(X_j)＝0,j＝1,…,N；a) The training data feature set X={X _j ∈ R ^K ,j=1,…,N} used for classifier training, and its corresponding identity Y={y _j ∈{0,1},j=1, ...,N}, where K=Ch×s is the number of features, where Ch is the number of channels, and s is the number of feature vectors extracted from a single lead in a single experiment, and N is the number of single leads contained in the training data. The number of experiments; F _m represents the classifier after m steps; the number of iterations is set as M; the EEG signal feature vector X _j of the jth single experiment is set as the initial probability P ₀ of imagining the tongue (y _j = 1|X _j )=0.5,j=1,...,N, set the initial classifier of the EEG signal feature vector X _j of the jth single experiment as F ₀ (X _j )=0,j=1, ..., N;

b)m表示迭代步数，从m＝1开始进行以下循环迭代：b) m represents the number of iteration steps, and the following loop iterations are performed from m=1:

i.求分类器F_m的似然函数的梯度

：i. Find the gradient of the likelihood function of the classifier F _m

:

${\overset{~ ~}{y the y}}_{j j} = = 22 (({y the y}_{j j} - - {P P}_{m m - - 11} (({y the y}_{j j} = = 11 | | {X x}_{j j})))),, j j = = 11,, . . . . . .,, N N$

其中，

为第m-1步迭代后，特征向量X_j属于想象舌头脑电的概率值；in,

After iterating for the m-1th step, the eigenvector X _j belongs to the probability value of the imagined tongue EEG;

ii.在最小二乘意义上，选择与梯度最相匹配的弱分类器f_m：ii. In the sense of least squares, select the weak classifier f _m that best matches the gradient:

${f f}_{m m} = = arg arg \underset{f f}{min min} {Σ Σ}_{j j = = 11}^{N N} {(({\overset{~ ~}{y the y}}_{j j} - - f f (({X x}_{j j}))))}^{22} = = {w w}^{T T} {X x}_{j j}$

其中，回归系数向量w由最小二乘算法求得。Among them, the regression coefficient vector w is obtained by the least squares algorithm.

iii.根据给定的训练数据得到F_m的伯努利对数似然函数：iii. Obtain the Bernoulli logarithmic likelihood function of F _m according to the given training data:

$L L (({F f}_{m m};; X x,, Y Y)) = = log log (({Π Π}_{j j = = 11}^{N N} {P P}_{m m} {(({y the y}_{j j} = = 11 | | {X x}_{j j}))}^{{y the y}_{j j}} {P P}_{m m} {(({y the y}_{j j} = = 00 | | {X x}_{j j}))}^{11 - - {y the y}_{j j}}));;$

iv.计算弱分类器f_m的权值γ_m为：iv. Calculate the weight γ _m of the weak classifier f _m as:

γ_m=argmaxL(F_m-1+γf_m;X,Y)；γ _m =argmaxL(F _m-1 +γ f _m ;X,Y);

v.更新分类器：v. Update the classifier:

F_m=F_m-1+εγ_mf_m；F _m =F _m-1 +εγ _m f _m ;

其中，ε为一极小值，设置为0.05；Among them, ε is a minimum value, set to 0.05;

vi.由分类器F_m计算特征向量X_j属于想象舌头脑电的概率值：vi. Calculate the probability value that the feature vector X _j belongs to the imagined tongue EEG by the classifier F _m :

${P P}_{m m} (({y the y}_{j j} = = 11 | | {X x}_{j j})) = = \frac{{e e}^{{F f}_{m m} (({X x}_{j j}))}}{{e e}^{{F f}_{m m} (({X x}_{j j}))} + + {e e}^{- - {F f}_{m m} (({X x}_{j j}))}},, j j = = 11,, . . . . . .,, N N$

其中，F_m(X_j)表示m步后对应训练数据X_j的分类器。Among them, F _m (X _j ) represents the classifier corresponding to the training data X _j after m steps.

vii.令m＝m+1，重复进行上述循环，如果m＝M，则循环迭代结束，得到的分类器F＝F_M；vii. Make m=m+1, repeat above-mentioned cycle, if m=M, then loop iteration ends, the classifier F=F _M that obtains;

步骤5）中所述的通过分类器计算输出概率值的方法为：将步骤4）中的盒维和分形截距特征向量X送入分类器F，利用公式：The method of calculating the output probability value through the classifier described in step 5) is: send the box dimension and fractal intercept feature vector X in step 4) to the classifier F, and use the formula:

$P P ((y the y = = 11 | | X x)) = = \frac{{e e}^{F f ((X x))}}{{e e}^{F f ((X x))} + + {e e}^{- - F f ((X x))}}$

得到脑电信号为想象舌头的概率P；Get the probability P that the EEG signal is an imagined tongue;

一种利用上述方法进行脑机接口的装置，包括以电路连接的脑电放大器、数据采集卡和计算机，所述计算机中内设置有检测脑电状态的脑电检测模块，利用脑电放大器和数据采集卡对脑电信号进行采集后传输到计算机中，利用脑电检测模块对脑电信号进行滤波和盒维、分形截距的特征提取，并将所提取的特征向量送入Boosting分类器中，获输出概率值；将输出概率值与预设阈值比较，得脑电状态检测结果并转化为控制外部设备的控制命令。A device for brain-computer interface using the above method, including a brain-electric amplifier connected with a circuit, a data acquisition card and a computer, the computer is provided with an brain-electricity detection module for detecting the state of the brain-electricity, using the brain-electricity amplifier and the data The acquisition card collects the EEG signal and transmits it to the computer, and uses the EEG detection module to filter the EEG signal and extract the features of the box dimension and fractal intercept, and send the extracted feature vector into the Boosting classifier. Obtain the output probability value; compare the output probability value with the preset threshold value to obtain the EEG state detection result and convert it into a control command for controlling the external device.

本发明的有益的效果是：The beneficial effects of the present invention are:

利用特征效果较好的盒维和分形截距对采集并经预处理后的脑电数据进行特征提取，将提取的特征向量送入Boosting分类器中，从而得到对不同想象运动的脑电信号的标记；在脑机接口技术领域中，本发明进一步提高了脑电信号分类正确率。Use the box dimension and fractal intercept with better feature effects to extract the features of the collected and preprocessed EEG data, and send the extracted feature vectors into the Boosting classifier, so as to obtain the marks of EEG signals for different imaginary movements ; In the field of brain-computer interface technology, the present invention further improves the classification accuracy of EEG signals.

附图说明Description of drawings

图1为本发明的结构框图；Fig. 1 is a block diagram of the present invention;

图2为本发明的流程框图；Fig. 2 is a block flow diagram of the present invention;

图3为滤波后想象左小指时的脑电信号；Figure 3 is the EEG signal when imagining the left little finger after filtering;

图4为滤波后想象舌头时的脑电信号；Fig. 4 is the EEG signal when imagining the tongue after filtering;

图5为脑电信号盒维特征的变化图；Fig. 5 is the change chart of EEG signal box-dimensional feature;

图6为脑电信号分形截距特征的变化图。Fig. 6 is a change diagram of the fractal intercept feature of the EEG signal.

具体实施方式Detailed ways

下面结合附图与实例对本发明做进一步说明，本发明并不限于此；The present invention will be further described below in conjunction with accompanying drawing and example, and the present invention is not limited thereto;

实施例1、Embodiment 1,

如图1-6所示；As shown in Figure 1-6;

本发明通过电极采集脑电信号，脑电信号经过脑电放大器放大及数据采集卡，输入计算机实现脑电信号分类，并产生控制命令控制外部设备；The invention collects EEG signals through electrodes, and the EEG signals are amplified by EEG amplifiers and data acquisition cards, and input into a computer to realize classification of EEG signals, and generate control commands to control external devices;

一种基于盒维和分形截距特征的脑机接口方法，其流程图如图2所示，包括以下步骤：A brain-computer interface method based on box dimension and fractal intercept features, the flow chart of which is shown in Figure 2, including the following steps:

1）采集实验者想象左小指、舌头运动时，大脑所产生的脑电信号，采样频率为1000Hz；实验者想象左小指运动时其脑电信号对应的类别标识为0类，想象舌头运动时其脑电信号对应的类别标识为1类，单次实验脑电信号时长为3秒；1) Collect the EEG signals generated by the brain when the experimenter imagines the movement of the left little finger and tongue, and the sampling frequency is 1000 Hz; The category corresponding to the EEG signal is identified as Category 1, and the duration of a single experimental EEG signal is 3 seconds;

采集的原始脑电信号如图3所示；选取实验者的前278次实验作为训练样本，其余的100次实验作为测试样本；The collected original EEG signals are shown in Figure 3; the first 278 experiments of the experimenter are selected as training samples, and the remaining 100 experiments are used as test samples;

3）对经步骤2）降采样后的脑电信号进行8-30Hz的带通滤波；所述的对脑电信号进行滤波的方法，包括步骤如下：3) performing 8-30 Hz band-pass filtering on the EEG signal after the downsampling in step 2); the method for filtering the EEG signal includes the following steps:

对脑电信号利用J阶的巴特沃斯滤波器进行带通滤波，优选J=5；滤波之后的脑电信号如图4所示；The EEG signal is band-pass filtered using a J-order Butterworth filter, preferably J=5; the EEG signal after filtering is shown in Figure 4;

a)将降采样和滤波后长为L=300点的脑电信号S，平均分割成长为H=100点的G=3个子段，然后计算每一个子段的盒维和分形截距；a) After downsampling and filtering, the EEG signal S with a length of L=300 points is averagely divided into G=3 sub-segments of H=100 points, and then the box dimension and fractal intercept of each sub-segment are calculated;

b)对脑电信号S的每个子段，将其继续分割成长为r的T个小段（r＝2^h,(0＜h＜log₂H)，T等于H/r的整数部分），取覆盖信号的盒子的边长等于r。对第i个小段（其中i＝1,…,T），若其振幅的最小值和最大值分别落在第k个和第l个盒子中，则覆盖第i个小段所需的盒子数n(i)为：b) For each sub-segment of the EEG signal S, continue to divide it into T small segments of r (r=2 ^h , (0<h<log ₂ H), T is equal to the integer part of H/r), take The side length of the box covering the signal is equal to r. For the i-th small segment (where i=1,...,T), if the minimum and maximum amplitudes fall in the k-th and l-th boxes respectively, then the number of boxes n required to cover the i-th small segment (i) is:

n(i)＝l-k+1；n(i)=l-k+1;

c)覆盖该子段所需的盒子总数Num(r)为c) The total number of boxes Num(r) required to cover the subsection is

$Num Num ((r r)) = = \underset{i i}{Σ Σ} n no ((i i))$

logNum(r)＝D·log(1r)+ClogNum(r)=D·log(1r)+C

其中，直线的斜率为D，截距为C；取不同的r值，计算得到若干组(r,Num(r))；应用最小二乘曲线拟合算法，求得该直线方程的斜率D和截距C；这里，斜率D即为该子段脑电信号的盒维数，而截距C则为该子段脑电信号的分形截距；图5为脑电信号的盒维特征，图6为脑电信号的分形截距特征；Among them, the slope of the straight line is D, and the intercept is C; different r values are taken to calculate several groups (r, Num(r)); the slope D and the slope of the straight line equation are obtained by applying the least squares curve fitting algorithm Intercept C; here, the slope D is the box dimension of the sub-segment EEG signal, and the intercept C is the fractal intercept of the sub-segment EEG signal; Fig. 5 is the box-dimensional feature of the EEG signal, Fig. 6 is the fractal intercept feature of the EEG signal;

步骤5）中所述的Boosting分类器的具体实现步骤为：The specific implementation steps of the Boosting classifier described in step 5) are:

a)分类器训练所使用的训练数据集X={X_j∈R^K,j=1,…，N}，其对应的标识Y={y_j∈{0,1},j=1,…,N}，其中，K=Ch×s是特征的数目，其中其中Ch是通道数等于64，而s是单次实验单个导联上所提取的特征向量的数目等于6，N为训练数据中所包含的单次实验的数目等于278；F_m表示m步后的分类器；设定迭代次数为M=200；设定第j次单次想象实验的脑电信号的特征向量X_j为想象舌头的初始概率P₀(y_j＝1|X_j)＝0.5,j＝1,…,N，设定第j次单次想象实验的脑电信号的特征向量X_j的初始分类器为F₀(X_j)＝0,j＝1,…,N；a) The training data set X={X _j ∈ R ^K ,j=1,…,N} used for classifier training, and its corresponding identification Y={y _j ∈{0,1},j=1,… ,N}, where K=Ch×s is the number of features, where Ch is the number of channels equal to 64, and s is the number of feature vectors extracted from a single lead in a single experiment equal to 6, and N is the number of features in the training data The number of included single experiments is equal to 278; F _m represents the classifier after m steps; the number of iterations is set to M=200; the feature vector X _j of the EEG signal of the j-th single imagination experiment is set as imagination The initial probability of the tongue P ₀ (y _j =1|X _j )=0.5,j=1,...,N, set the initial classifier of the feature vector X _j of the EEG signal of the j-th single imagination experiment as F ₀ (X _j )=0,j=1,...,N;

i.求分类器F_m的似然函数的梯度

：i. Find the gradient of the likelihood function of the classifier F _m

:

其中，

$L L (({F f}_{m m};; X x,, Y Y)) = = log log (({Π Π}_{j j = = 11}^{N N} {P P}_{m m} {(({y the y}_{j j} = = 11 | | {X x}_{j j}))}^{{y the y}_{j j}} {P P}_{m m} {(({y the y}_{j j} = = 00 | | {X x}_{j j}))}^{11 - - {y the y}_{j j}}))$

γ_m=argmaxL(F_m-1+γf_m;X,Y)γ _m =argmaxL(F _m-1 +γf _m ;X,Y)

v.更新分类器：v. Update the classifier:

F_m=F_m-1+εγ_mf_m；F _m =F _m-1 +εγ _m f _m ;

当输出概率值小于或等于预设阈值时，则判断此时的脑电状态为想象左小指时的脑电信号，并转换为控制命令2。When the output probability value is less than or equal to the preset threshold, it is judged that the EEG state at this time is the EEG signal when imagining the left little finger, and it is converted into a control command 2 .

实施例2、Embodiment 2,

一种利用如实施例1所述方法进行脑机接口的装置，如图2所示，包括以电路连接的脑电放大器、数据采集卡和计算机，所述计算机中内设置有检测脑电状态的脑电检测模块，利用脑电放大器和数据采集卡对脑电信号进行采集后传输到计算机中，利用脑电检测模块对脑电信号进行滤波和盒维、分形截距的特征提取，并将所提取的特征向量送入Boosting分类器中，获输出概率值；将输出概率值与预设阈值比较，得脑电状态检测结果并转化为控制轮椅的控制命令。A kind of device that utilizes the method described in embodiment 1 to carry out brain-computer interface, as shown in Figure 2, comprises the brain electric amplifier, data acquisition card and computer connected with the circuit, is provided with and detects brain electric state in the described computer The EEG detection module uses the EEG amplifier and the data acquisition card to collect the EEG signals and transmits them to the computer, and uses the EEG detection module to filter the EEG signals and extract the features of box dimension and fractal intercept. The extracted feature vector is sent to the Boosting classifier to obtain the output probability value; the output probability value is compared with the preset threshold value to obtain the EEG state detection result and convert it into a control command to control the wheelchair.

利用本发明对测试脑电样本进行检测，识别的正确率达92%。The invention is used to detect the test EEG samples, and the correct rate of recognition reaches 92%.

Claims

1. A brain-computer interface method based on box dimension and fractal intercept features, characterized in that the method comprises the following steps:

1) Collect the EEG signals generated by the brain when the experimenter imagines the movement of the left little finger and tongue, and the sampling frequency is 1000 Hz; The category corresponding to the EEG signal is identified as category 1;

2) Down-sampling the collected EEG signal, the sampling frequency is 100Hz;

3) Perform 8-30Hz band-pass filtering on the EEG signal after step 2) downsampling;

4) Extract the box dimension and fractal intercept features of each channel in the EEG signal, wherein the steps for extracting the box dimension and fractal intercept of each channel of the EEG signal are:

a) The down-sampled and filtered EEG signal S with a length of L points is averagely divided into G sub-segments of H points, and then the box dimension and fractal intercept of each sub-segment are calculated;

b) For each sub-segment of the EEG signal S, continue to divide it into T small segments of r (r=2 ^h , (0<h<log ₂ H), T is equal to the integer part of H/r), take The side length of the box covering the signal is equal to r. For the i-th small segment (where i=1,...,T), if the minimum and maximum amplitudes fall in the k-th and l-th boxes respectively, then the number of boxes n required to cover the i-th small segment (i) is:

n(i)=l-k+1;

c) The total number of boxes Num(r) required to cover this subsection is:

d) The box dimension D of the sub-segment EEG signal is:

e) When the side length r of the box changes, the formula described in step d) satisfies the straight line equation:

logNum(r)＝D·log(1r)+C

Among them, the slope of the straight line is D, and the intercept is C; different r values are taken to calculate several groups (r, Num(r)); the slope D and the slope of the straight line equation are obtained by applying the least squares curve fitting algorithm Intercept C; here, the slope D is the box dimension of the sub-segment EEG signal, and the intercept C is the fractal intercept of the sub-segment EEG signal;

5) Input the box dimension and fractal intercept features extracted in step 4) to the Boosting classifier for classification, and obtain the output probability value;

6) Compare the output probability value with the preset threshold value, wherein the preset threshold value is 0.5, obtain the EEG state detection result and convert it into a control command:

When the output probability value is greater than the preset threshold, it is judged that the EEG state at this time is the EEG signal when the tongue is imagined, and converted into a control command 1;

When the output probability value is less than or equal to the preset threshold, it is judged that the EEG state at this time is the EEG signal when imagining the left little finger, and it is converted into a control command 2 . the

2. A brain-computer interface method based on box dimension and fractal intercept features according to claim 1, characterized in that the method for filtering EEG signals described in step 3) comprises the following steps:

The EEG signal is band-pass filtered at 8-30 Hz using a J-order Butterworth filter, preferably J=5. the

3. according to claim 1, step 5) in the realization step of the Boosting classifier is:

a) The training data set X={X _j ∈ R ^K ,j=1,…,N} used for classifier training, and its corresponding identification Y={y _j ∈{0,1},j=1,… , N}, where K=Ch×s is the number of features, where Ch is the number of channels equal to 64, and s is the number of feature vectors extracted from a single lead in a single experiment equal to 6, and N is the number of features in the training data The number of included single experiments is equal to 278; F _m represents the classifier after m steps; the number of iterations is set to M=200; the feature vector X _j of the EEG signal of the j-th single imagination experiment is set as imagination The initial probability of the tongue P ₀ (y _j =1|X _j )=0.5,j=1,...,N, set the initial classifier of the feature vector X _j of the EEG signal of the j-th single imagination experiment as F ₀ (X _j )=0,j=1,...,N;

b) m represents the number of iteration steps, and the following loop iterations are performed starting from m=1:

i. Find the gradient of the likelihood function of the classifier F _m :

in,

ii. In the sense of least squares, select the weak classifier f _m that best matches the gradient:

Among them, the regression coefficient vector w is obtained by the least squares algorithm. the

iii. Obtain the Bernoulli logarithmic likelihood function of F _m according to the given training data:

iv. Calculate the weight γ _m of the weak classifier f _m as:

γ _m =argmaxL(F _m-1 +γ f _m ;X,Y);

v. Update the classifier:

F _m =F _m-1 +εγ _m f _m ;

Among them, ε is a minimum value, set to 0.05;

vi. Calculate the probability value that the feature vector X _j belongs to the imagined tongue EEG by the classifier F _m :

Among them, F _m (X _j ) represents the classifier corresponding to the training data X _j after m steps.

vii. Let m=m+1, repeat the above cycle, if m=M, the cycle iteration ends, and the obtained classifier F=F _M .

4. A brain-computer interface method based on box dimension and fractal intercept features according to claim 1, characterized in that the method of calculating the output probability value through the classifier described in step 5) is:

Send the box dimension and fractal intercept feature vector X in step 4) to the classifier F, using the formula:

Get the probability P that the EEG signal is the imagined tongue. the

5. A device utilizing the method as claimed in claim 1 to carry out a brain-computer interface, characterized in that the device comprises a circuit-connected EEG amplifier, a data acquisition card, and a computer, and a device for detecting EEG is arranged in the computer. The EEG detection module of the state uses the EEG amplifier and the data acquisition card to collect the EEG signal and transmits it to the computer, and uses the EEG detection module to filter the EEG signal and extract the features of box dimension and fractal intercept, and Send the extracted feature vector into the Boosting classifier to obtain the output probability value; compare the output probability value with the preset threshold value to obtain the EEG state detection result and convert it into a control command to control the external device. the