一种基于改进ResNet的疲劳检测方法

A Fatigue Detection Method Based on Improved ResNet

为了避免因疲劳驾驶导致的交通事故的发生,该文提出一种基于改进ResNet网络的疲劳检测方法。通过在残差模块中引入深度卷积,对标准化方法进行优化并引入平均池化和注意机制,有效提升了模型对脑电信号全局信息的捕捉能力。通过SEED-VIG的功率谱特征和微分熵特征数据集、多熵融合疲劳检测数据集和SPIS静息状态数据集对该文模型进行验证。实验结果表明,该文模型在各数据集上的平均准确率达到97.4%,较原ResNet网络提升17.9%。此模型对疲劳检测十分有效,可在一定程度上降低交通事故发生率。

In order to avoid traffic accidents caused by tired driving,a fatigue detection method based on improved ResNet network was proposed.By introducing deep convolution into the residual module,the standardization method is optimized,and the average pooling and attention mechanism are introduced to effectively improve the model’s ability to capture the global information of EEG signal.The SEED-VIG power spectrum feature and differential entropy feature dataset,multi-entropy fusion fatigue detection dataset and SPIS resting state dataset were used to verify the proposed model.The experimental results show that the average accuracy of the proposed model on each data set is 97.4%,which is 17.9%higher than that of the original ResNet network.This model is very effective for the detection of fatigue state and can reduce the incidence of traffic accidents to a certain extent.