面向智能驾培系统的深度学习表情识别方法

Deep learning facial expression recognition method for intelligent driving training system

针对虚拟现实驾培系统人机交互不足、不能很好地替代教练员的作用的问题,提出将基于迁移学习的表情识别算法嵌入驾培系统,使系统可以根据学员的情绪状态智能调整培训程序、提供个性化学习方案。首先,对人脸表情训练数据集做了归一化处理和数据增强处理,引入ReLU激活函数,利用ImageNet数据集在改进型的EfficientNet上进行预训练,使网络具有良好的特征提取能力;然后将预训练得到的网络模型进行微调,获得最终识别模型。通过Fer2013数据集和CK+数据集实验验证,改进后的EfficientNet鲁棒性更好,可以满足虚拟现实驾培系统对学员面部表情的实时识别需求。

Aiming at the problem that human-computer interaction of virtual reality driving training system is not enough to replace the role of coaches, propose to embed the facial expression recognition algorithm based on transfer learning into the driving training system, so that the system can adjust the training program intelligently according to the students’ emotional state and provide personalized learning programs.Firstly, the face expression training dataset is normalizing and data enhancing processing, ReLu activation function is introduced, and ImageNet dataset is used to carry out pre-training on the improved EfficientNet, so as to make the network has good feature extraction ability.Then the network model obtained by pre-training is fine-tuned to obtain the final recognition model.The improved EfficientNet EfficientNet is verified by Fer2013 dataset and CK+ dataset experiments, which can meet the real-time recognition requirements of virtual reality driving training system for students’ facial expressions.