重庆理工大学学报2025，Vol.39Issue(5)：34-42,9.DOI:10.3969/j.issn.1674-8425(z).2025.03.005
融合深度残差网络与注意力机制的驾驶人行为检测方法研究
Study on driver behavior detection method combining deep residual networks and attention mechanism
摘要
Abstract
To improve the accuracy of driver behavior detection and the model interpretability,we propose a driver behavior detection model that integrates deep residual networks with attention mechanisms.First,the advantages of the feature extraction module of the deep residual network are utilized to compare the results of network models with different numbers of layers.Then,an appropriate network model is selected as the base network.Next,to eliminate the interference of irrelevant information on driving behavior,the SE Block attention mechanism is introduced to perform feature extraction and classification prediction on images.Finally,the performance of the proposed model is validated through comparative experiments with other models,ablation studies,and feature visualization experiments.Our results show the average classification accuracy of our model is 99.89%,superior to that of other detection models.The Grad-CAM visualization method is employed to explain our model's focus areas,enabling it to focus more precisely on the key features of driving behaviors,thereby enhancing its interpretability and thus helping it earn more people's confidence.关键词
深度学习/驾驶人行为检测/深度残差网络/注意力机制/神经网络可视化Key words
deep learning/driver behavior surveillance/deep residual network/attention mechanism/neural network visualization分类
计算机与自动化引用本文复制引用
陈运星,崔军华,吴钊,吴华伟,袁星宇..融合深度残差网络与注意力机制的驾驶人行为检测方法研究[J].重庆理工大学学报,2025,39(5):34-42,9.基金项目
湖北省自然科学基金创新发展联合基金项目(2024AFD045) （2024AFD045）
襄阳市研究与开发项目(2022ABH006510) （2022ABH006510）
