發(fā)布時間：2018-02-28 18:08

  本文關鍵詞： 腦電圖(EEG) 徑向基神經(jīng)網(wǎng)絡 直推式遷移學習　出處：《計算機科學與探索》2016年12期 　論文類型：期刊論文

【摘要】：在癲癇腦電圖(electroencephalogram,EEG)信號識別中,傳統(tǒng)的智能建模方法要求訓練數(shù)據(jù)集和測試數(shù)據(jù)集均服從相同的分布。但在實際應用中,某些情況并不能滿足此條件,進而導致傳統(tǒng)方法性能急劇下降。針對上述情況,引入遷移學習策略,提出了適用于數(shù)據(jù)分布遷移環(huán)境的直推式徑向基神經(jīng)網(wǎng)絡(transductive radial basis function neural network,TRBFNN)。該方法在癲癇EEG信號識別中的實驗結果表明:直推式徑向基神經(jīng)網(wǎng)絡具有較好的場景遷移適應性,對訓練數(shù)據(jù)和測試數(shù)據(jù)存在差異時,識別性能不會出現(xiàn)急劇惡化的現(xiàn)象。
[Abstract]:In the recognition of EEG electroencephalogramma (EGG) signals, the traditional intelligent modeling method requires that both the training data set and the test data set be distributed in the same way. However, in some practical applications, this condition cannot be satisfied. Therefore, the performance of the traditional methods drops sharply. In view of the above situation, the transfer learning strategy is introduced. A direct radial basis function neural network (TRBFNN) for data distribution migration environment is proposed. The experimental results of this method in EEG signal recognition show that the direct push radial basis function neural network has a good adaptability to scene migration. When there are differences between training data and test data, the recognition performance will not deteriorate sharply.
【作者單位】： 江南大學數(shù)字媒體技術學院;
【基金】：國家自然科學基金面上項目No.61170122 江蘇省杰出青年基金項目No.BK20140001 新世紀優(yōu)秀人才支持計劃項目No.NCET-120882~~
【分類號】：R742.1;TP183
，

本文編號：1548375