一种预测驾驶注意力的多尺度注意力模型  

A Multi-Scale Attention Model for Driving Attention Prediction

作　　者：蒋超 郜东瑞 李芃锐 赵长名 JIANG Chao;GAO Dongrui;LI Pengrui;ZHAO Changming(School of Computer Science,Chengdu University of Information Science and Technology,Chengdu 610225,China)

机构地区：[1]成都信息工程大学计算机学院,四川成都610225

出　　处：《软件导刊》2025年第4期18-24,共7页Software Guide

基　　金：国家重点研发计划项目(2021YFF1200605);LOST 2030脑科学项目(2022ZD0208500);成都信息工程大学科研基金项目(KYQN202208、KYQN202206);四川省科技计划项目(2023NSFSC0499);国家自然科学基金项目(62272067)。

摘　　要：近年来,许多研究致力于利用EEG与EOG多模态数据预测驾驶注意力,但有效融合这两种模态数据仍是一项充满挑战的任务。为此,提出一个基于多模态的多尺度通道注意力回归模型(MMCAR-Net)预测驾驶注意力。首先,通过多尺度感知单元(Multi-Scale Inception)从EEG、EOG模态数据中分别提取多尺度特征;其次,在多个尺度上有序合并EEG与EOG特征以增强融合特征的多样性;最后,引入多尺度通道注意力机制为多尺度特征赋予差异化权重,以强化与注意力预测相关的特征,提升模型对驾驶注意力相关特征的敏感性和表达能力。在SEED-VIG数据集上的实验表明,所提模型在个体内实验组中取得的PCC与RMSE分别为0.959和0.064,在跨被试实验组中对应数值为0.892和0.112。In recent years,many studies have been dedicated to the prediction of driving attention by utilizing multimodal data derived from EEG and EOG.However,effectively fusing these two modal data remains a challenging task.In this paper,a methodology named the multi‐modal multiscale channel attention regression network for driving attention prediction was introduced.The model employs a multi-scale incep‐tion unit to extract multi-scale features from both EEG and EOG data.Subsequently,it sequentially merges EEG and EEG features at different scales,enhancing the diversity of fused features.Concurrently,a multi-scale channel attention mechanism is introduced to assign differentiat‐ed weights to multi-scale features to strengthen the features related to attention prediction,and to improve the model's sensitivity and expres‐siveness to driving attention-related features.Through a series of experiments conducted on the SEED-VIG dataset,the proposed model achieved a PCC of 0.959 and RMSE of 0.064 in the intra-individual experimental group,and corresponding values of 0.892 and 0.112 in the cross-subjects experimental group,showcasing exceptional performance.

关 键 词：驾驶注意力预测 多尺度感知单元 多尺度通道注意力机制 特征融合 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]