Abstract: Intelligent recognition of electroencephalogram (EEG) signals is an important means of epilepsy detection. Almost all the traditional intelligent recognition methods assume that the training samples and the testing samples of EEG signals are drawn from same distribution. However in many practical applications for epilepsy detection, the assumption is too harsh, and it is invalid when there are some differences between the distribution of the training and testing data. In order to overcome this challenge, this paper introduces the widely concerned transfer learning method into EEG signals recognition. With the transfer learning based method, the distribution between the training and testing data is not required to be the same, which expands the application scene of the algorithm and realizes the adaptive recognition for EEG signals in the data drift scene. The experimental results show that the new method for epilepsy detection has a better adaptability than the traditional methods.

Key words: electroencephalogram (EEG), wavelet transform, epilepsy detection, transfer learning, feature extraction

摘要： 脑电图（electroencephalogram，EEG）信号智能识别是癫痫病检测的重要手段。传统的智能识别方法在进行检测时，都假定智能模型训练采用的训练样本集和测试样本集满足同一分布特征，但在实际应用时，此假设条件过于苛刻，当训练和测试数据对应的场景有一定漂移时传统方法不再适用。针对上述情况，将近年来广受关注的对分布差异性场景具备较好性能的迁移学习方法引入到脑电图识别中，使得最终所得的模型对训练和测试数据的分布要求较之传统方法得到进一步放松，扩大了算法的适应场景，实现了在数据漂移场景下对癫痫EEG信号的自适应识别。实验表明，基于迁移学习的方法比传统方法具有更好的适应性。

关键词: 脑电图（EEG）, 小波变换, 癫痫识别, 迁移学习, 特征提取