{"id":90,"date":"2026-02-13T10:29:54","date_gmt":"2026-02-13T02:29:54","guid":{"rendered":"https:\/\/qdszxyy.com\/?p=90"},"modified":"2026-04-29T07:47:48","modified_gmt":"2026-04-28T23:47:48","slug":"%e8%84%91%e6%9c%ba%e6%8e%a5%e5%8f%a3%e6%95%b0%e6%8d%ae%e5%a4%84%e7%90%86%e8%bf%9e%e8%bd%bd%ef%bc%88%e4%b8%80%ef%bc%89%e5%85%a5%e9%97%a8%ef%bc%9a%e6%a0%b8%e5%bf%83%e6%a6%82%e5%bf%b5%e4%b8%8e%e6%8a%80","status":"publish","type":"post","link":"https:\/\/qdszxyy.com\/?p=90","title":{"rendered":"\u8111\u673a\u63a5\u53e3\u6570\u636e\u5904\u7406\u8fde\u8f7d\uff08\u4e00\uff09\u5165\u95e8\uff1a\u6838\u5fc3\u6982\u5ff5\u4e0e\u6280\u672f\u6846\u67b6(\u8f6c\u8f7d)"},"content":{"rendered":"

\u8111\u673a\u63a5\u53e3\uff08Brain-Computer Interface, BCI\uff09\u662f\u4e00\u79cd\u4ee4\u4eba\u5174\u594b\u7684\u6280\u672f\uff0c\u5b83\u80fd\u591f\u5c06\u5927\u8111\u6d3b\u52a8\u76f4\u63a5\u8f6c\u6362\u4e3a\u8ba1\u7b97\u673a\u6307\u4ee4\uff0c\u4e3a\u6b8b\u969c\u4eba\u58eb\u63d0\u4f9b\u65b0\u7684\u4ea4\u6d41\u65b9\u5f0f\uff0c\u4e5f\u4e3a\u6e38\u620f\u3001\u533b\u7597\u7b49\u9886\u57df\u5e26\u6765\u9769\u547d\u6027\u7684\u53d8\u5316\u3002\u672c\u6587\u5c06\u5e26\u60a8\u8d70\u8fdb BCI \u6570\u636e\u5904\u7406\u7684\u4e16\u754c\uff0c\u4ece\u6838\u5fc3\u6982\u5ff5\u5230\u6280\u672f\u6846\u67b6\uff0c\u518d\u5230\u5b9e\u9645\u4ee3\u7801\u5b9e\u73b0\uff0c\u5e2e\u52a9\u60a8\u5feb\u901f\u5165\u95e8\u8fd9\u4e00\u524d\u6cbf\u9886\u57df\u3002<\/p>\n

1. \u6838\u5fc3\u6982\u5ff5
\n1.1 \u8111\u7535\u4fe1\u53f7\uff08EEG\uff09\u57fa\u7840
\n\u8111\u7535\u56fe\uff08Electroencephalography, EEG\uff09\u662f BCI \u6700\u5e38\u7528\u7684\u6570\u636e\u6765\u6e90\uff0c\u5b83\u901a\u8fc7\u653e\u7f6e\u5728\u5934\u76ae\u4e0a\u7684\u7535\u6781\u8bb0\u5f55\u5927\u8111\u7684\u7535\u6d3b\u52a8\u3002EEG \u4fe1\u53f7\u5177\u6709\u4ee5\u4e0b\u7279\u70b9\uff1a<\/p>\n

\u9891\u7387\u8303\u56f4\uff1a0.5-100 Hz
\n\u632f\u5e45\u8303\u56f4\uff1a1-100 \u03bcV
\n\u9ad8\u65f6\u95f4\u5206\u8fa8\u7387\uff08\u6beb\u79d2\u7ea7\uff09
\n\u4f4e\u7a7a\u95f4\u5206\u8fa8\u7387
\n\u4e3b\u8981\u7684 EEG \u4fe1\u53f7\u9891\u6bb5\u5305\u62ec\uff1a<\/p>\n

\u03b4 \u6ce2\uff080.5-4 Hz\uff09\uff1a\u6df1\u5ea6\u7761\u7720
\n\u03b8 \u6ce2\uff084-8 Hz\uff09\uff1a\u56f0\u5026\u3001\u653e\u677e
\n\u03b1 \u6ce2\uff088-13 Hz\uff09\uff1a\u6e05\u9192\u4f46\u653e\u677e
\n\u03b2 \u6ce2\uff0813-30 Hz\uff09\uff1a\u8b66\u89c9\u3001\u4e13\u6ce8
\n\u03b3 \u6ce2\uff08>30 Hz\uff09\uff1a\u9ad8\u7ea7\u8ba4\u77e5\u529f\u80fd
\n1.2 \u5e38\u89c1 BCI \u8303\u5f0f
\n\u6839\u636e\u4fe1\u53f7\u7c7b\u578b\u548c\u63a7\u5236\u65b9\u5f0f\uff0cBCI \u4e3b\u8981\u5206\u4e3a\u4ee5\u4e0b\u51e0\u7c7b\uff1a<\/p>\n

\u8fd0\u52a8\u60f3\u8c61\uff08Motor Imagery, MI\uff09BCI\uff1a\u7528\u6237\u901a\u8fc7\u60f3\u8c61\u80a2\u4f53\u8fd0\u52a8\uff08\u5982\u5de6\u624b\u3001\u53f3\u624b\u3001\u811a\u7684\u8fd0\u52a8\uff09\u6765\u4ea7\u751f\u7279\u5b9a\u7684\u8111\u7535\u4fe1\u53f7\u6a21\u5f0f\u3002<\/p>\n

P300 BCI\uff1a\u57fa\u4e8e\u4e8b\u4ef6\u76f8\u5173\u7535\u4f4d\uff08ERP\uff09\u4e2d\u7684 P300 \u6210\u5206\uff0c\u5f53\u7528\u6237\u6ce8\u610f\u5230\u7279\u5b9a\u523a\u6fc0\u65f6\uff0c\u5728\u523a\u6fc0\u51fa\u73b0\u7ea6 300ms \u540e\u4f1a\u51fa\u73b0\u4e00\u4e2a\u6b63\u5411\u7535\u4f4d\u3002<\/p>\n

\u7a33\u6001\u89c6\u89c9\u8bf1\u53d1\u7535\u4f4d\uff08SSVEP\uff09BCI\uff1a\u7528\u6237\u6ce8\u89c6\u4e0d\u540c\u9891\u7387\u95ea\u70c1\u7684\u89c6\u89c9\u523a\u6fc0\uff0c\u5927\u8111\u4f1a\u4ea7\u751f\u4e0e\u523a\u6fc0\u9891\u7387\u540c\u6b65\u7684\u8111\u7535\u4fe1\u53f7\u3002<\/p>\n

\u76ae\u5c42\u8111\u7535\uff08ECoG\uff09BCI\uff1a\u7535\u6781\u76f4\u63a5\u653e\u7f6e\u5728\u5927\u8111\u76ae\u5c42\u8868\u9762\uff0c\u63d0\u4f9b\u66f4\u9ad8\u8d28\u91cf\u7684\u4fe1\u53f7\uff0c\u4f46\u9700\u8981\u624b\u672f\u690d\u5165\u3002<\/p>\n

1.3 BCI \u6570\u636e\u5904\u7406\u6d41\u7a0b
\n\u4e00\u4e2a\u5b8c\u6574\u7684 BCI \u7cfb\u7edf\u901a\u5e38\u5305\u62ec\u4ee5\u4e0b\u51e0\u4e2a\u5904\u7406\u9636\u6bb5\uff1a<\/p>\n

\u6570\u636e\u91c7\u96c6\uff1a\u901a\u8fc7 EEG \u8bbe\u5907\u8bb0\u5f55\u8111\u7535\u4fe1\u53f7
\n\u9884\u5904\u7406\uff1a\u53bb\u9664\u566a\u58f0\u548c artifacts\uff0c\u589e\u5f3a\u4fe1\u53f7\u8d28\u91cf
\n\u7279\u5f81\u63d0\u53d6\uff1a\u4ece\u9884\u5904\u7406\u540e\u7684\u4fe1\u53f7\u4e2d\u63d0\u53d6\u5177\u6709\u5224\u522b\u6027\u7684\u7279\u5f81
\n\u5206\u7c7b \/ \u89e3\u7801\uff1a\u4f7f\u7528\u673a\u5668\u5b66\u4e60\u7b97\u6cd5\u5c06\u7279\u5f81\u6620\u5c04\u5230\u7279\u5b9a\u7684\u7528\u6237\u610f\u56fe
\n\u5e94\u7528\u8f93\u51fa\uff1a\u5c06\u89e3\u7801\u7ed3\u679c\u8f6c\u6362\u4e3a\u5b9e\u9645\u7684\u5e94\u7528\u63a7\u5236\u6307\u4ee4
\n2. \u6280\u672f\u6846\u67b6\u8be6\u89e3
\n2.1 \u6570\u636e\u91c7\u96c6
\n\u6570\u636e\u91c7\u96c6\u662f BCI \u7cfb\u7edf\u7684\u57fa\u7840\uff0c\u9700\u8981\u4e13\u4e1a\u7684 EEG \u8bbe\u5907\u3002\u5e38\u89c1\u7684\u8bbe\u5907\u5305\u62ec\uff1a<\/p>\n

\u4f20\u7edf\u6709\u7ebf EEG \u7cfb\u7edf\uff08\u5982 Neuroscan, Biosemi\uff09
\n\u4fbf\u643a\u5f0f\u65e0\u7ebf EEG \u8bbe\u5907\uff08\u5982 Emotiv, Muse\uff09
\n\u7814\u7a76\u7ea7\u9ad8\u5bc6\u5ea6 EEG \u7cfb\u7edf\uff0864\/128\/256 \u901a\u9053\uff09
\n\u6570\u636e\u91c7\u96c6\u7684\u5173\u952e\u53c2\u6570\uff1a<\/p>\n

\u91c7\u6837\u7387\uff1a\u901a\u5e38 250-1000 Hz
\n\u7535\u6781\u6570\u91cf\uff1a\u4ece\u5c11\u6570\u51e0\u4e2a\u5230\u6570\u767e\u4e2a\u4e0d\u7b49
\n\u53c2\u8003\u7535\u6781\u4f4d\u7f6e
\n\u63a5\u5730\u7535\u6781\u4f4d\u7f6e
\n2.2 \u9884\u5904\u7406
\n\u9884\u5904\u7406\u662f BCI \u6570\u636e\u5904\u7406\u4e2d\u81f3\u5173\u91cd\u8981\u7684\u4e00\u6b65\uff0c\u76ee\u7684\u662f\u53bb\u9664\u566a\u58f0\u548c artifacts\uff0c\u63d0\u9ad8\u4fe1\u53f7\u8d28\u91cf\u3002\u5e38\u89c1\u7684\u9884\u5904\u7406\u6b65\u9aa4\u5305\u62ec\uff1a<\/p>\n

\u6ee4\u6ce2\uff1a<\/p>\n

\u9ad8\u901a\u6ee4\u6ce2\uff1a\u53bb\u9664\u4f4e\u9891\u6f02\u79fb\uff08\u901a\u5e38 0.5-1 Hz\uff09
\n\u4f4e\u901a\u6ee4\u6ce2\uff1a\u53bb\u9664\u9ad8\u9891\u566a\u58f0\uff08\u901a\u5e38 30-50 Hz\uff09
\n\u9677\u6ce2\u6ee4\u6ce2\uff1a\u53bb\u9664\u5de5\u9891\u5e72\u6270\uff08\u5982 50 Hz \u6216 60 Hz\uff09
\n\u4f2a\u8ff9\u53bb\u9664\uff1a<\/p>\n

\u773c\u7535\u4f2a\u8ff9\uff08EOG\uff09\uff1a\u7531\u773c\u775b\u8fd0\u52a8\u4ea7\u751f
\n\u808c\u7535\u4f2a\u8ff9\uff08EMG\uff09\uff1a\u7531\u808c\u8089\u6d3b\u52a8\u4ea7\u751f
\n\u5fc3\u7535\u4f2a\u8ff9\uff08ECG\uff09\uff1a\u7531\u5fc3\u810f\u6d3b\u52a8\u4ea7\u751f
\n\u65b9\u6cd5\uff1a\u72ec\u7acb\u6210\u5206\u5206\u6790\uff08ICA\uff09\u3001\u81ea\u9002\u5e94\u6ee4\u6ce2\u3001\u9608\u503c\u53bb\u9664\u7b49
\n\u4fe1\u53f7\u6807\u51c6\u5316\uff1a<\/p>\n

\u5f52\u4e00\u5316\uff1a\u5c06\u4fe1\u53f7\u7f29\u653e\u5230\u7279\u5b9a\u8303\u56f4
\n\u6807\u51c6\u5316\uff1a\u4f7f\u4fe1\u53f7\u5177\u6709\u96f6\u5747\u503c\u548c\u5355\u4f4d\u65b9\u5dee
\n2.3 \u7279\u5f81\u63d0\u53d6
\n\u7279\u5f81\u63d0\u53d6\u662f\u5c06\u539f\u59cb EEG \u4fe1\u53f7\u8f6c\u6362\u4e3a\u66f4\u5177\u5224\u522b\u6027\u7684\u7279\u5f81\u5411\u91cf\u7684\u8fc7\u7a0b\u3002\u5e38\u89c1\u7684\u7279\u5f81\u63d0\u53d6\u65b9\u6cd5\u5305\u62ec\uff1a<\/p>\n

\u65f6\u57df\u7279\u5f81\uff1a<\/p>\n

\u5747\u503c\u3001\u65b9\u5dee\u3001\u5cf0\u503c\u3001\u8c37\u503c
\n\u8fc7\u96f6\u7387
\n\u6ce2\u5f62\u957f\u5ea6
\n\u81ea\u56de\u5f52\u6a21\u578b\u7cfb\u6570\uff08AR\uff09
\n\u9891\u57df\u7279\u5f81\uff1a<\/p>\n

\u529f\u7387\u8c31\u5bc6\u5ea6\uff08PSD\uff09
\n\u7279\u5b9a\u9891\u6bb5\u7684\u529f\u7387\uff08\u03b1, \u03b2, \u03b8, \u03b4, \u03b3\uff09
\n\u9891\u8c31\u71b5
\n\u5c0f\u6ce2\u53d8\u6362\u7cfb\u6570
\n\u65f6\u9891\u57df\u7279\u5f81\uff1a<\/p>\n

\u77ed\u65f6\u5085\u91cc\u53f6\u53d8\u6362\uff08STFT\uff09
\n\u5c0f\u6ce2\u53d8\u6362\uff08WT\uff09
\n\u5e0c\u5c14\u4f2f\u7279 – \u9ec4\u53d8\u6362\uff08HHT\uff09
\n\u7a7a\u95f4\u7279\u5f81\uff1a<\/p>\n

\u8111\u7535\u5730\u5f62\u56fe
\n\u4e0d\u540c\u7535\u6781\u95f4\u7684\u76f8\u5173\u6027
\n\u5171\u540c\u7a7a\u95f4\u6a21\u5f0f\uff08CSP\uff09
\n2.4 \u5206\u7c7b \/ \u89e3\u7801
\n\u5206\u7c7b\u662f\u5c06\u63d0\u53d6\u7684\u7279\u5f81\u6620\u5c04\u5230\u7528\u6237\u610f\u56fe\u7684\u8fc7\u7a0b\u3002\u5e38\u7528\u7684\u5206\u7c7b\u7b97\u6cd5\u5305\u62ec\uff1a<\/p>\n

\u4f20\u7edf\u673a\u5668\u5b66\u4e60\u7b97\u6cd5\uff1a<\/p>\n

\u652f\u6301\u5411\u91cf\u673a\uff08SVM\uff09
\n\u7ebf\u6027\u5224\u522b\u5206\u6790\uff08LDA\uff09
\n\u4eba\u5de5\u795e\u7ecf\u7f51\u7edc\uff08ANN\uff09
\nk \u8fd1\u90bb\uff08k-NN\uff09
\n\u51b3\u7b56\u6811\u548c\u968f\u673a\u68ee\u6797
\n\u6df1\u5ea6\u5b66\u4e60\u65b9\u6cd5\uff1a<\/p>\n

\u5377\u79ef\u795e\u7ecf\u7f51\u7edc\uff08CNN\uff09
\n\u5faa\u73af\u795e\u7ecf\u7f51\u7edc\uff08RNN\uff09
\n\u957f\u77ed\u671f\u8bb0\u5fc6\u7f51\u7edc\uff08LSTM\uff09
\n\u6df1\u5ea6\u5b66\u4e60\u4e0e\u4f20\u7edf\u7279\u5f81\u63d0\u53d6\u7ed3\u5408
\n\u8bc4\u4f30\u6307\u6807\uff1a<\/p>\n

\u51c6\u786e\u7387\uff08Accuracy\uff09
\n\u6df7\u6dc6\u77e9\u9635\uff08Confusion Matrix\uff09
\nkappa \u7cfb\u6570
\n\u4fe1\u606f\u4f20\u8f93\u7387\uff08ITR\uff09
\n2.5 \u5e94\u7528\u8f93\u51fa
\n\u5e94\u7528\u8f93\u51fa\u5c06\u5206\u7c7b\u7ed3\u679c\u8f6c\u6362\u4e3a\u5b9e\u9645\u7684\u63a7\u5236\u6307\u4ee4\uff0c\u5e38\u89c1\u7684\u5e94\u7528\u5305\u62ec\uff1a<\/p>\n

\u8f6e\u6905\u63a7\u5236
\n\u5047\u80a2\u63a7\u5236
\n\u6587\u5b57\u8f93\u5165
\n\u6e38\u620f\u63a7\u5236
\n\u795e\u7ecf\u53cd\u9988\u6cbb\u7597
\n3. \u5b9e\u6218\u4ee3\u7801\u793a\u4f8b
\n\u4e0b\u9762\u6211\u4eec\u5c06\u901a\u8fc7\u4e00\u4e2a\u5b8c\u6574\u7684\u4ee3\u7801\u793a\u4f8b\u6765\u6f14\u793a BCI \u6570\u636e\u5904\u7406\u7684\u57fa\u672c\u6d41\u7a0b\u3002\u6211\u4eec\u5c06\u4f7f\u7528 MNE \u5e93\u5904\u7406\u4e00\u4e2a\u516c\u5f00\u7684 BCI \u6570\u636e\u96c6\u3002<\/p>\n

3.1 \u5b89\u88c5\u5fc5\u8981\u7684\u5e93
\npython<\/p>\n

\u8fd0\u884c<\/p>\n

!pip install mne numpy scipy matplotlib scikit-learn
\n\u4e00\u952e\u83b7\u53d6\u5b8c\u6574\u9879\u76ee\u4ee3\u7801
\n3.2 \u6570\u636e\u52a0\u8f7d\u4e0e\u63a2\u7d22
\n\u6211\u4eec\u5c06\u4f7f\u7528 BCI Competition IV \u7684\u6570\u636e\u96c6 2a\uff0c\u8fd9\u662f\u4e00\u4e2a\u8fd0\u52a8\u60f3\u8c61 BCI \u6570\u636e\u96c6\u3002<\/p>\n

python<\/p>\n

\u8fd0\u884c<\/p>\n

import mne
\nimport numpy as np
\nimport matplotlib.pyplot as plt
\nfrom sklearn.preprocessing import StandardScaler
\nfrom sklearn.discriminant_analysis import LinearDiscriminantAnalysis
\nfrom sklearn.model_selection import cross_val_score, train_test_split
\nfrom sklearn.pipeline import Pipeline
\nfrom sklearn.svm import SVC
\nfrom sklearn.metrics import classification_report, confusion_matrix<\/p>\n

# \u52a0\u8f7d\u793a\u4f8b\u6570\u636e
\n# \u6ce8\u610f\uff1a\u8fd9\u91cc\u6211\u4eec\u4f7f\u7528MNE\u5185\u7f6e\u7684\u793a\u4f8b\u6570\u636e\uff0c\u5b9e\u9645\u4f7f\u7528\u65f6\u9700\u8981\u4e0b\u8f7d\u5b8c\u6574\u6570\u636e\u96c6
\nfrom mne.datasets import bcic4_2a
\nraw_fname, event_fname = bcic4_2a.data_path()<\/p>\n

# \u8bfb\u53d6\u539f\u59cb\u6570\u636e
\nraw = mne.io.read_raw_edf(raw_fname, preload=True)
\nevents = mne.read_events(event_fname)<\/p>\n

# \u67e5\u770b\u6570\u636e\u57fa\u672c\u4fe1\u606f
\nprint(“\u6570\u636e\u5f62\u72b6:”, raw.get_data().shape)
\nprint(“\u91c7\u6837\u7387:”, raw.info[‘sfreq’], “Hz”)
\nprint(“\u7535\u6781\u6570\u91cf:”, len(raw.info[‘ch_names’]))
\nprint(“\u7535\u6781\u540d\u79f0:”, raw.info[‘ch_names’])<\/p>\n

# \u67e5\u770b\u4e8b\u4ef6\u4fe1\u606f
\nprint(“\\n\u4e8b\u4ef6\u7c7b\u578b:”, np.unique(events[:, 2]))
\nprint(“\u4e8b\u4ef6\u6570\u91cf:”, len(events))<\/p>\n

# \u7ed8\u5236\u539f\u59cb\u6570\u636e\u7684\u524d10\u79d2
\nraw.plot(duration=10, n_channels=8, scalings=’auto’)
\nplt.show()
\n\u4e00\u952e\u83b7\u53d6\u5b8c\u6574\u9879\u76ee\u4ee3\u7801<\/p>\n

3.3 \u6570\u636e\u9884\u5904\u7406
\npython<\/p>\n

\u8fd0\u884c<\/p>\n

# \u8bbe\u7f6e\u53c2\u8003\u7535\u6781
\nraw.set_eeg_reference(‘average’, projection=True)<\/p>\n

# \u6ee4\u6ce2\uff1a\u53bb\u9664\u76f4\u6d41\u5206\u91cf\u548c\u5de5\u9891\u5e72\u6270
\nraw.filter(l_freq=0.5, h_freq=30, fir_design=’firwin’)<\/p>\n

# \u9677\u6ce2\u6ee4\u6ce2\u53bb\u966450Hz\u5de5\u9891\u5e72\u6270
\nraw.notch_filter(freqs=50, fir_design=’firwin’)<\/p>\n

# \u63d0\u53d6\u4e8b\u4ef6\u76f8\u5173\u7684\u65f6\u95f4\u6bb5
\nevent_id = {
\n‘left_hand’: 1,
\n‘right_hand’: 2,
\n‘foot’: 3,
\n‘tongue’: 4
\n}
\ntmin, tmax = -0.2, 0.5 # \u4ece\u523a\u6fc0\u524d200ms\u5230\u523a\u6fc0\u540e500ms<\/p>\n

# \u521b\u5efaepochs
\nepochs = mne.Epochs(raw, events, event_id, tmin, tmax,
\nproj=True, baseline=(None, 0), preload=True)<\/p>\n

print(“Epochs\u5f62\u72b6:”, epochs.get_data().shape) # (n_epochs, n_channels, n_times)<\/p>\n

# \u53ef\u89c6\u5316\u4e0d\u540c\u7c7b\u522b\u7684\u4e8b\u4ef6\u76f8\u5173\u7535\u4f4d
\nepochs[‘left_hand’].average().plot()
\nepochs[‘right_hand’].average().plot()
\nplt.show()
\n\u4e00\u952e\u83b7\u53d6\u5b8c\u6574\u9879\u76ee\u4ee3\u7801<\/p>\n

3.4 \u7279\u5f81\u63d0\u53d6
\n\u6211\u4eec\u5c06\u63d0\u53d6\u65f6\u57df\u548c\u9891\u57df\u7279\u5f81\uff1a<\/p>\n

python<\/p>\n

\u8fd0\u884c<\/p>\n

# \u63d0\u53d6\u7279\u5f81\uff1a\u65f6\u57df\u7279\u5f81\u548c\u9891\u57df\u7279\u5f81
\ndef extract_features(epochs):
\nn_epochs, n_channels, n_times = epochs.get_data().shape
\nfeatures = []<\/p>\n

for i in range(n_epochs):
\nepoch_data = epochs.get_data()[i]<\/p>\n

# \u65f6\u57df\u7279\u5f81\uff1a\u5747\u503c\u3001\u65b9\u5dee\u3001\u5cf0\u503c\u3001\u8c37\u503c
\nmean = np.mean(epoch_data, axis=1)
\nvar = np.var(epoch_data, axis=1)
\npeak = np.max(np.abs(epoch_data), axis=1)
\ntrough = np.min(epoch_data, axis=1)<\/p>\n

# \u9891\u57df\u7279\u5f81\uff1a\u4f7f\u7528\u529f\u7387\u8c31\u5bc6\u5ea6
\npsd, freqs = mne.time_frequency.psd_array_welch(
\nepoch_data, sfreq=epochs.info[‘sfreq’], fmin=0.5, fmax=30, n_fft=128
\n)<\/p>\n

# \u63d0\u53d6\u4e0d\u540c\u9891\u6bb5\u7684\u5e73\u5747\u529f\u7387
\nalpha_power = np.mean(psd[:, (freqs >= 8) & (freqs <= 13)], axis=1)
\nbeta_power = np.mean(psd[:, (freqs >= 13) & (freqs <= 30)], axis=1)<\/p>\n

# \u7ec4\u5408\u6240\u6709\u7279\u5f81
\nepoch_features = np.concatenate([mean, var, peak, trough, alpha_power, beta_power])
\nfeatures.append(epoch_features)<\/p>\n

return np.array(features)<\/p>\n

# \u63d0\u53d6\u7279\u5f81
\nX = extract_features(epochs)
\ny = epochs.events[:, 2] # \u4e8b\u4ef6\u6807\u7b7e<\/p>\n

print(“\u7279\u5f81\u5f62\u72b6:”, X.shape)
\nprint(“\u6807\u7b7e\u5f62\u72b6:”, y.shape)
\nprint(“\u7c7b\u522b\u5206\u5e03:”, np.bincount(y))
\n\u4e00\u952e\u83b7\u53d6\u5b8c\u6574\u9879\u76ee\u4ee3\u7801<\/p>\n

3.5 \u5206\u7c7b\u4e0e\u8bc4\u4f30
\npython<\/p>\n

\u8fd0\u884c<\/p>\n

# \u5212\u5206\u8bad\u7ec3\u96c6\u548c\u6d4b\u8bd5\u96c6
\nX_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42)<\/p>\n

# \u521b\u5efa\u9884\u5904\u7406\u548c\u5206\u7c7b\u7684\u6d41\u6c34\u7ebf
\npipeline = Pipeline([
\n(‘scaler’, StandardScaler()), # \u6807\u51c6\u5316
\n(‘lda’, LinearDiscriminantAnalysis()) # LDA\u5206\u7c7b\u5668
\n])<\/p>\n

# \u4f7f\u7528\u4ea4\u53c9\u9a8c\u8bc1\u8bc4\u4f30\u6a21\u578b
\ncv_scores = cross_val_score(pipeline, X_train, y_train, cv=5)
\nprint(“\u4ea4\u53c9\u9a8c\u8bc1\u51c6\u786e\u7387: {:.2f} \u00b1 {:.2f}”.format(cv_scores.mean(), cv_scores.std()))<\/p>\n

# \u5728\u6d4b\u8bd5\u96c6\u4e0a\u8bc4\u4f30
\npipeline.fit(X_train, y_train)
\ny_pred = pipeline.predict(X_test)
\naccuracy = np.mean(y_pred == y_test)
\nprint(“\u6d4b\u8bd5\u96c6\u51c6\u786e\u7387:”, accuracy)<\/p>\n

# \u6253\u5370\u5206\u7c7b\u62a5\u544a
\nprint(“\\n\u5206\u7c7b\u62a5\u544a:”)
\nprint(classification_report(y_test, y_pred, target_names=[‘\u4f11\u606f’, ‘\u5de6\u624b’, ‘\u53f3\u624b’, ‘\u811a’, ‘\u820c\u5934’]))<\/p>\n

# \u7ed8\u5236\u6df7\u6dc6\u77e9\u9635
\ncm = confusion_matrix(y_test, y_pred)
\nplt.imshow(cm, interpolation=’nearest’, cmap=plt.cm.Blues)
\nplt.title(‘\u6df7\u6dc6\u77e9\u9635’)
\nplt.colorbar()
\ntick_marks = np.arange(len(event_id))
\nplt.xticks(tick_marks, event_id.keys(), rotation=45)
\nplt.yticks(tick_marks, event_id.keys())
\nplt.tight_layout()
\nplt.ylabel(‘\u771f\u5b9e\u6807\u7b7e’)
\nplt.xlabel(‘\u9884\u6d4b\u6807\u7b7e’)
\nplt.show()
\n\u4e00\u952e\u83b7\u53d6\u5b8c\u6574\u9879\u76ee\u4ee3\u7801<\/p>\n

3.6 \u4f7f\u7528 CSP \u7279\u5f81\u7684\u8fd0\u52a8\u60f3\u8c61\u5206\u7c7b
\n\u5171\u540c\u7a7a\u95f4\u6a21\u5f0f\uff08CSP\uff09\u662f\u8fd0\u52a8\u60f3\u8c61 BCI \u4e2d\u5e38\u7528\u7684\u7279\u5f81\u63d0\u53d6\u65b9\u6cd5\uff1a<\/p>\n

python<\/p>\n

\u8fd0\u884c<\/p>\n

from mne.decoding import CSP<\/p>\n

# \u4e3a\u4e86\u7b80\u5316\uff0c\u6211\u4eec\u53ea\u9009\u62e9\u5de6\u624b\u548c\u53f3\u624b\u7684\u8fd0\u52a8\u60f3\u8c61\u6570\u636e
\nepochs_mi = epochs[‘left_hand’, ‘right_hand’]
\nX_mi = epochs_mi.get_data()
\ny_mi = epochs_mi.events[:, 2]<\/p>\n

# \u5212\u5206\u8bad\u7ec3\u96c6\u548c\u6d4b\u8bd5\u96c6
\nX_train_mi, X_test_mi, y_train_mi, y_test_mi = train_test_split(
\nX_mi, y_mi, test_size=0.2, random_state=42
\n)<\/p>\n

# \u521b\u5efaCSP\u7279\u5f81\u63d0\u53d6\u5668\u548c\u5206\u7c7b\u5668\u7684\u6d41\u6c34\u7ebf
\ncsp = CSP(n_components=4, reg=’ledoit_wolf’, log=True)
\nlda = LinearDiscriminantAnalysis()<\/p>\n

# \u8bad\u7ec3CSP
\ncsp.fit(X_train_mi, y_train_mi)<\/p>\n

# \u63d0\u53d6CSP\u7279\u5f81
\nX_train_csp = csp.transform(X_train_mi)
\nX_test_csp = csp.transform(X_test_mi)<\/p>\n

# \u8bad\u7ec3LDA\u5206\u7c7b\u5668
\nlda.fit(X_train_csp, y_train_mi)<\/p>\n

# \u8bc4\u4f30
\naccuracy_csp = lda.score(X_test_csp, y_test_mi)
\nprint(“CSP+LDA\u5206\u7c7b\u51c6\u786e\u7387:”, accuracy_csp)<\/p>\n

# \u53ef\u89c6\u5316CSP\u6a21\u5f0f
\ncsp.plot_patterns(epochs_mi.info, ch_type=’eeg’, units=’Patterns (AU)’, size=1.5)
\nplt.show()<\/p>\n

# \u53ef\u89c6\u5316CSP\u8fc7\u6ee4\u540e\u7684\u4fe1\u53f7
\nepochs_filt = csp.transform(epochs_mi)
\nplt.figure(figsize=(12, 6))
\nfor i in range(4):
\nplt.subplot(2, 2, i+1)
\nplt.plot(epochs_filt[y_mi==1, i].mean(axis=0), label=’\u5de6\u624b’)
\nplt.plot(epochs_filt[y_mi==2, i].mean(axis=0), label=’\u53f3\u624b’)
\nplt.title(f’CSP\u6210\u5206 {i+1}’)
\nplt.legend()
\nplt.show()
\n\u4e00\u952e\u83b7\u53d6\u5b8c\u6574\u9879\u76ee\u4ee3\u7801<\/p>\n

3.7 \u6df1\u5ea6\u5b66\u4e60\u65b9\u6cd5\u793a\u4f8b
\n\u4f7f\u7528 CNN \u5904\u7406 EEG \u6570\u636e\uff1a<\/p>\n

python<\/p>\n

\u8fd0\u884c<\/p>\n

import tensorflow as tf
\nfrom tensorflow.keras.models import Sequential
\nfrom tensorflow.keras.layers import Conv2D, MaxPooling2D, Flatten, Dense, Dropout<\/p>\n

# \u51c6\u5907CNN\u8f93\u5165\u6570\u636e (n_samples, n_channels, n_times, 1)
\nX_cnn = X_mi[:, :, :, np.newaxis]<\/p>\n

# \u5212\u5206\u8bad\u7ec3\u96c6\u548c\u6d4b\u8bd5\u96c6
\nX_train_cnn, X_test_cnn, y_train_cnn, y_test_cnn = train_test_split(
\nX_cnn, y_mi, test_size=0.2, random_state=42
\n)<\/p>\n

# \u5c06\u6807\u7b7e\u8f6c\u6362\u4e3a\u72ec\u70ed\u7f16\u7801
\ny_train_onehot = tf.keras.utils.to_categorical(y_train_cnn – 1, num_classes=2)
\ny_test_onehot = tf.keras.utils.to_categorical(y_test_cnn – 1, num_classes=2)<\/p>\n

# \u521b\u5efaCNN\u6a21\u578b
\nmodel = Sequential()
\nmodel.add(Conv2D(32, (1, 5), activation=’relu’, input_shape=(X_train_cnn.shape[1], X_train_cnn.shape[2], 1)))
\nmodel.add(MaxPooling2D((1, 2)))
\nmodel.add(Conv2D(64, (1, 5), activation=’relu’))
\nmodel.add(MaxPooling2D((1, 2)))
\nmodel.add(Flatten())
\nmodel.add(Dense(128, activation=’relu’))
\nmodel.add(Dropout(0.5))
\nmodel.add(Dense(2, activation=’softmax’))<\/p>\n

# \u7f16\u8bd1\u6a21\u578b
\nmodel.compile(optimizer=’adam’,
\nloss=’categorical_crossentropy’,
\nmetrics=[‘accuracy’])<\/p>\n

# \u8bad\u7ec3\u6a21\u578b
\nhistory = model.fit(X_train_cnn, y_train_onehot,
\nepochs=20, batch_size=16,
\nvalidation_split=0.2)<\/p>\n

# \u8bc4\u4f30\u6a21\u578b
\ntest_loss, test_acc = model.evaluate(X_test_cnn, y_test_onehot)
\nprint(“CNN\u6d4b\u8bd5\u51c6\u786e\u7387:”, test_acc)<\/p>\n

# \u7ed8\u5236\u8bad\u7ec3\u5386\u53f2
\nplt.figure(figsize=(12, 4))
\nplt.subplot(1, 2, 1)
\nplt.plot(history.history[‘loss’], label=’\u8bad\u7ec3\u635f\u5931’)
\nplt.plot(history.history[‘val_loss’], label=’\u9a8c\u8bc1\u635f\u5931’)
\nplt.xlabel(‘Epoch’)
\nplt.ylabel(‘Loss’)
\nplt.legend()<\/p>\n

plt.subplot(1, 2, 2)
\nplt.plot(history.history[‘accuracy’], label=’\u8bad\u7ec3\u51c6\u786e\u7387’)
\nplt.plot(history.history[‘val_accuracy’], label=’\u9a8c\u8bc1\u51c6\u786e\u7387’)
\nplt.xlabel(‘Epoch’)
\nplt.ylabel(‘Accuracy’)
\nplt.legend()
\nplt.show()
\n\u4e00\u952e\u83b7\u53d6\u5b8c\u6574\u9879\u76ee\u4ee3\u7801<\/p>\n

4. \u6311\u6218\u4e0e\u672a\u6765\u53d1\u5c55
\n\u5c3d\u7ba1 BCI \u6280\u672f\u53d6\u5f97\u4e86\u663e\u8457\u8fdb\u5c55\uff0c\u4f46\u4ecd\u9762\u4e34\u8bb8\u591a\u6311\u6218\uff1a<\/p>\n

\u4fe1\u53f7\u8d28\u91cf\uff1aEEG \u4fe1\u53f7\u5fae\u5f31\u4e14\u6613\u53d7\u566a\u58f0\u5e72\u6270
\n\u4e2a\u4f53\u5dee\u5f02\uff1a\u4e0d\u540c\u4eba\u7684\u8111\u7535\u4fe1\u53f7\u6a21\u5f0f\u5b58\u5728\u5f88\u5927\u5dee\u5f02
\n\u957f\u671f\u7a33\u5b9a\u6027\uff1aBCI \u7cfb\u7edf\u7684\u6027\u80fd\u53ef\u80fd\u968f\u65f6\u95f4\u53d8\u5316
\n\u7528\u6237\u8bad\u7ec3\uff1a\u7528\u6237\u901a\u5e38\u9700\u8981\u5927\u91cf\u8bad\u7ec3\u624d\u80fd\u719f\u7ec3\u4f7f\u7528 BCI \u7cfb\u7edf
\n\u7a7a\u95f4\u5206\u8fa8\u7387\uff1a\u5934\u76ae EEG \u7684\u7a7a\u95f4\u5206\u8fa8\u7387\u8f83\u4f4e
\n\u672a\u6765\u7684\u53d1\u5c55\u65b9\u5411\u5305\u62ec\uff1a<\/p>\n

\u9ad8\u5206\u8fa8\u7387 EEG\uff1a\u589e\u52a0\u7535\u6781\u6570\u91cf\uff0c\u63d0\u9ad8\u7a7a\u95f4\u5206\u8fa8\u7387
\n\u6df7\u5408 BCI\uff1a\u7ed3\u5408\u591a\u79cd\u4fe1\u53f7\u6a21\u6001\uff08EEG, EOG, EMG \u7b49\uff09
\n\u95ed\u73af\u7cfb\u7edf\uff1a\u5b9e\u65f6\u53cd\u9988\u548c\u81ea\u9002\u5e94\u8c03\u6574
\n\u6df1\u5ea6\u5b66\u4e60\u5e94\u7528\uff1a\u7aef\u5230\u7aef\u7684 BCI \u7cfb\u7edf
\n\u975e\u4fb5\u5165\u6027\u8111\u673a\u63a5\u53e3\u7684\u7a81\u7834\uff1a\u65e0\u9700\u624b\u672f\u5373\u53ef\u83b7\u5f97\u9ad8\u8d28\u91cf\u4fe1\u53f7
\n5. \u603b\u7ed3\u4e0e\u8d44\u6e90\u63a8\u8350
\n\u672c\u6587\u4ecb\u7ecd\u4e86\u8111\u673a\u63a5\u53e3\u6570\u636e\u5904\u7406\u7684\u6838\u5fc3\u6982\u5ff5\u548c\u6280\u672f\u6846\u67b6\uff0c\u5305\u62ec\u6570\u636e\u91c7\u96c6\u3001\u9884\u5904\u7406\u3001\u7279\u5f81\u63d0\u53d6\u3001\u5206\u7c7b\u548c\u89e3\u7801\u7b49\u5173\u952e\u6b65\u9aa4\u3002\u901a\u8fc7\u5b9e\u9645\u4ee3\u7801\u793a\u4f8b\uff0c\u5c55\u793a\u4e86\u5982\u4f55\u4f7f\u7528 MNE \u5e93\u548c\u673a\u5668\u5b66\u4e60\u65b9\u6cd5\u5904\u7406 BCI \u6570\u636e\u3002<\/p>\n

\u2014\u2014\u2014\u2014\u2014\u2014\u2014\u2014\u2014\u2014\u2014\u2014\u2014\u2014\u2014\u2014
\n\u7248\u6743\u58f0\u660e\uff1a\u672c\u6587\u4e3aCSDN\u535a\u4e3b\u300cBrduino\u8111\u673a\u63a5\u53e3\u6280\u672f\u7b54\u7591\u300d\u7684\u539f\u521b\u6587\u7ae0\uff0c\u9075\u5faaCC 4.0 BY-SA\u7248\u6743\u534f\u8bae\uff0c\u8f6c\u8f7d\u8bf7\u9644\u4e0a\u539f\u6587\u51fa\u5904\u94fe\u63a5\u53ca\u672c\u58f0\u660e\u3002
\n\u539f\u6587\u94fe\u63a5\uff1ahttps:\/\/blog.csdn.net\/m0_63827302\/article\/details\/155226004<\/p>\n

<\/div>\n

<\/path><\/svg><\/i> \"Loading\"<\/p>\n

<\/div>\n","protected":false},"excerpt":{"rendered":"

\u8111\u673a\u63a5\u53e3\uff08Brain-Computer Interface, BCI\uff09\u662f\u4e00\u79cd\u4ee4\u4eba\u5174\u594b\u7684\u6280\u672f\uff0c\u5b83\u80fd\u591f\u5c06\u5927\u8111\u6d3b\u52a8\u76f4 […]<\/p>\n

<\/div>\n

<\/path><\/svg><\/i> \"Loading\"<\/p>\n

<\/div>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[13,11,21,14],"tags":[],"class_list":["post-90","post","type-post","status-publish","format-standard","hentry","category-bci_","category-bci","category-bci_apple","category-eeg"],"_links":{"self":[{"href":"https:\/\/qdszxyy.com\/index.php?rest_route=\/wp\/v2\/posts\/90","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/qdszxyy.com\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/qdszxyy.com\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/qdszxyy.com\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/qdszxyy.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=90"}],"version-history":[{"count":1,"href":"https:\/\/qdszxyy.com\/index.php?rest_route=\/wp\/v2\/posts\/90\/revisions"}],"predecessor-version":[{"id":91,"href":"https:\/\/qdszxyy.com\/index.php?rest_route=\/wp\/v2\/posts\/90\/revisions\/91"}],"wp:attachment":[{"href":"https:\/\/qdszxyy.com\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=90"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/qdszxyy.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=90"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/qdszxyy.com\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=90"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}