<p>Performance indicators using CWT.</p>

To improve the detection performance of epileptic electroencephalogram (EEG) signals and address their non-stationary characteristics,this paper compares the combined effects of continuous wavelet transform (CWT) and short-time Fourier transform (STFT) with three neural network models—EEGNet,AlexNet,and Shallow ConvNet—and incorporates targeted optimization designs. Specifically,Focal Loss,dynamic data augmentation,and an early stopping mechanism are introduced in the training phase to enhance model robustness. For EEGNet,optimizations are implemented by integrating a Squeeze-and-Excitation (SE) attention module,improving depthwise separable convolution,and dynamically adapting dimensions to reduce classification errors. For Shallow ConvNet,improvements include layered convolution for extracting “time-frequency” features and average pooling to adapt to long-duration data blocks. Experiments are conducted based on subject-independent validation,and the results show that the CWT-based feature extraction method outperforms STFT comprehensively. Among all combinations,the CWT+Shallow ConvNet pair exhibits the optimal overall performance,while the CWT+EEGNet combination follows closely with excellent precision. These findings verify the effectiveness of combining precise time-frequency features (extracted by CWT) with optimized neural network models,providing reliable technical support for clinical epileptic EEG signal detection.为提升癫痫脑电图（electroencephalogram, EEG）信号的检测性能并解决其非平稳特性，本文针对连续小波变换（continuous wavelet transform, CWT）与短时傅里叶变换（short-time Fourier transform, STFT）分别结合三种神经网络模型——EEGNet、AlexNet及Shallow ConvNet——的联合效果开展对比研究，并引入针对性优化设计。具体而言，本文在训练阶段引入焦点损失（Focal Loss）、动态数据增强及早停机制，以提升模型鲁棒性。针对EEGNet，本文通过集成压缩与激励（Squeeze-and-Excitation, SE）注意力模块、优化深度可分离卷积结构及动态适配维度的方式实现优化，以降低分类误差。针对Shallow ConvNet，其优化方案包括采用分层卷积提取「时频」特征，以及使用平均池化适配长时长数据块。本文基于受试者独立验证开展实验，结果表明基于CWT的特征提取方法全面优于STFT方法。在所有组合方案中，CWT+Shallow ConvNet的组合展现出最优的综合性能，而CWT+EEGNet的组合紧随其后，具备优异的分类精度。上述研究结果验证了将CWT提取的精准时频特征与优化后的神经网络模型相结合的有效性，可为临床癫痫脑电图信号检测提供可靠的技术支撑。