基于知识蒸馏的心律失常分类模型

Arrhythmia classification model based on knowledge distillation

近些年ECG心律失常自动分类技术发展迅速,使用深度神经网络对心律失常进行自动分类取得了较好的分类效果,但同时深度神经网络也有着计算量大、占用内存多等缺点,难以部署在资源受限的小型设备上。为了降低运用在ECG心律失常分类上的神经网络模型的参数量和计算量,并使模型保持较高的分类精度,提出一种基于知识蒸馏和注意力机制的轻量级卷积神经网络模型。通过将美国麻省理工学院提供的研究心律失常的MIT-BIH数据库中的数据进行数据增强后生成图像数据集,对模型进行训练和测试。模型在保持较小的参数量及计算量的同时,在心律失常分类测试中达到了98.3%的分类准确率。

In recent years,ECG arrhythmia automatic classification technology has developed rapidly.The use of deep neural networks to automatically classify arrhythmia has achieved good classification results.At the same time,deep neural networks also have disadvantages such as large calculations and large memory usage,making it difficult to deploy on small devices with limited resources.In order to reduce the amount of parameters and calculations of the neural network model used in ECG arrhythmia classification,and to maintain the model’s high classification accuracy,a lightweight convolutional neural network model based on knowledge distillation and attention mechanism is proposed.The image data set is generated by data enhancement of the data in the MIT-BIH database,and the model is trained and tested.While maintaining a small amount of parameters and calculations,the model achieved a classification accuracy of 98.3%in the arrhythmia classification test.