基于改进ReviewKD的心电图分类模型研究

Study on ECG Classification Model Based on Improved ReviewKD

针对边缘计算环境中心电图分类模型精度难以进一步提升的问题,本文提出一种基于改进ReviewKD的心电图分类模型。首先在ReviewKD的注意力融合结构中引入反卷积层和GAM注意力机制,使学生模型具备更强的学习能力和全局特征提取能力;然后在损失函数中加入非目标类归一化损失函数,增强学生模型的抗噪声能力。在MIT-BIH数据集上进行实验,结果表明,改进ReviewKD的学生模型分类性能佳,适合用作边缘计算环境的心电图分类模型。

In order to improve the accuracy of ECG classification model in edge computing environment,an improved ReviewKD based ECG classification model is proposed in this paper.Firstly,the deconvolution layer and GAM attention mechanism are introduced into the attention fusion structure of ReviewKD to make the student model have stronger learning ability and global feature extraction ability.Then the non-target class normalized loss function is added to the loss function to enhance the anti-noise ability of the student model.Experiments on MIT-BIH dataset show that the modified ReviewKD student model has good classification performance and is suitable for ECG classification in edge computing environment.