基于语音特征迁移学习的驾驶疲劳检测

针对现有方法在实际应用时的标记样本稀缺与测试样本数据分布偏移等问题,提出一种基于语音特征迁移学习的驾驶疲劳检测方法。通过基于迁移学习的特征空间变换,对源领域有标记样本与目标域无标记样本数据间的边缘分布、条件分布、流形结构进行联合适配及降维处理,以解决样本数据分布偏移和特征维度过高的问题。以半监督学习的方式来迭代优化目标域样本的伪标记,并据此不断更新特征变换方式和迁移分类器,进而提高疲劳检测模型的精度和泛化能力。通过实验将本文方法与现有常用的监督学习、半监督学习和迁移学习等方法进行对比。结果表明,在测试时间、应用场景和被试个体均发生变化的情况下,本文所提方法的驾驶疲劳检测效果显著优于现有方法,正确率最高达到86.7%,具有实际应用价值。

Driver Fatigue Detection Based on Speech Feature Transfer Learning

Aiming at the scarcity of labeled samples and the data distribution deviation of test samples in practical application,a driver fatigue detection method based on speech feature transfer learning was proposed.Firstly,through the feature space transformation based on transfer learning,the marginal distribution,conditional distribution and manifold structure between the labeled sample data in the source domain and the unlabeled samples data in the target domain were jointly adapted and dimensionality reduced,in order to solve the problem of the sample data distribution deviation and high feature dimension.Secondly,the pseudo-labels of the target domain samples were iteratively optimized by using the semi-supervised learning method.Meanwhile,the feature transformation mode and the transfer classifier were updated synchronously to improve the accuracy and generalization ability of the fatigue detection model.At last,experiments were conducted to compare the proposed method with the existing commonly used methods,such as supervised learning,semi-supervised learning and transfer learning.The results show that,when the test time,application scenarios and subjects are changed,the driving fatigue detection effect of this method is significantly better than the existing methods,with highest accuracy rate of 86.7%,which is of practical value.