无人驾驶中基于ResNet深度模型的人脸微表情识别算法.pdf

第 1 期 微 处 理 机 No. 1 2023 年 2 月 MICROPROCESSORS Feb. ，2023 无人驾驶中基于 ResNet深度模型的人脸 微表情识别算法 * 蔡 臻，李 锋，魏楚强 （广东交通职业技术学院信息学院，广州510650） 摘 要: 为解决由于人脸微表情存在局部特征少、不同情绪差异性小等特点而带来的特征提取 难度大、表情识别率低等问题，以进一步提高人脸微表情识别精度，提出一种基于ResNet深度模型的 人脸微表情识别算法。算法针对无人驾驶中的行人人脸表情识别场合，主要包括数据预处理及模型构 建，在保证数据集统一性的同时，能够有效提高微表情的识别率。使用FER2013数据集对ResNet-50 模型进行实际验证，并与ResNet-18 的表现加以对比。本算法在实验中获得98.7%的准确率，优于 ResNet-18，充分验证算法模型的有效 。 ： ；无人驾驶 关键词 ResNet模型；微表情；人脸识别 DOI：10.3969/j.issn.1002-2279.2023.01.009 中图分类号:TP391.41 文献标识码:A 文章编号：1002-2279（2023）01-0036-04 FacialMicro-Expression RecognitionAlgorithm Basedon ResNetDepthModelinUnmannedDriving CAIZhen, LI Feng,WEI Chuqiang (SchoolofInformation, GuangdongCommunicationPolytechnic, Guangzhou510650, China) Abstract: Inorder to solvethe problems of difficultfeature extraction andlow expression recognition rate caused by the lack of local features and small differencesbetween different emotions in facial micro- expressions, and to further improve the accuracy of facial micro-expressions, a facial micro-expression recognition algorithm based on ResNet depth model is proposed. The algorithm mainly includes data pre- processing and model construction,which can effectivelyimprovethe recognitionrate of micro-expressions while ensuring the uniformity of data sets. The FER2013 data set is used to verify the ResNet-50 model, and the performance is compared with that of ResNet-18. The accuracy of the algorithm is 98.7% in the experiment,which isbetterthan ResNet-18,fully verifyingthe effectivenessofthe algorithmmodel. Keywords: ResNetmodel; Microexpression;