基于可调Q因子小波变换和迁移学习的癫痫脑电信号检测

针对癫痫脑电信号的检测问题,提出一种基于可调Q因子小波变换和迁移学习的癫痫脑电信号检测方法。首先,对EEG信号进行可调Q因子小波变换,并选择能量差异较大的子带进行部分重构,重排重构信号,将其表示为二维彩色图像数据;其次,通过对现有的癫痫发作自动检测算法和深度可分离卷积网络Xception模型的分析,使用ImageNet数据集分类的预训练模型参数进行网络参数初始化,得到深度可分离卷积网络Xception的预训练模型;最后,利用迁移学习方法将Xception模型的预训练结果迁移至癫痫发作自动检测任务。所提方法在BONN癫痫数据集上的准确度达到99.37%,敏感度达到100%,特异度达到98.48%,证明了该模型在癫痫发作自动检测任务上具有良好的泛化能力。与传统检测方法和其他深度学习方法相比,所提自动检测方法达到了较高的准确率,避免了人工设计和提取特征的过程,具有较好的应用价值。

Epileptic EEG Signals Detection Based on Tunable Q-factor Wavelet Transform and Transfer Learning

Aiming at the detection of epileptic EEG signals,a method of detecting epileptic EEG signals based on Tunable Q-factor wavelet transform and transfer learning is proposed.Firstly,the EEG signals are transformed by Tunable Q-factor wavelet transform,and the subbands with large energy differences are selected for partial reconstruction.The reconstructed signals are rearranged and expressed as two-dimensional color image data.Secondly,through the analysis of the existing automatic seizure detection algorithm and the Xception model of deep separable convolutional networks,the parameters of the pre-training model classified by the ImageNet dataset are used to initialize the network parameters,and the pre-training model of the depth separable convolution network Xception is obtained.Finally,the transfer learning method is used to transfer the pre-training results of the Xception model to the automatic seizure detection task.The performance of this method is verified on the BONN epilepsy dataset,and the accuracy,sensitivity and specificity reaches 99.37%,100%and 98.48%respectively,proving that the model has good generalization ability in automatic seizure detection task.Compared with traditional detection methods and other deep lear-ning methods based,the automatic detection method proposed in this paper achieves higher accuracy,avoids the process of artificial design and feature extraction,and has better application value.