融合注意力机制与残差网络的人耳识别方法  

Ear recognition method combining attention mechanism and residual network

作　　者：曹淑欣 许学斌 路龙宾 刘晨光 CAO Shuxin;XU Xuebin;LU Longbin;LIU Chenguang(School of Computer Science and Technology,Xi'an University of Posts&Telecommunications,Xi'an,Shaanxi 710121,China;Shannxi Key Laboratory of Network Data Analysis and Intelligent Processing,Xi'an University of Posts&Telecommunications,Xi'an,Shaanxi 710121,China)

机构地区：[1]西安邮电大学计算机学院,陕西西安710121 [2]西安邮电大学陕西省网络数据分析与智能处理重点实验室,陕西西安710121

出　　处：《光电子．激光》2023年第4期378-386,共9页Journal of Optoelectronics·Laser

基　　金：国家自然科学基金(61673316);陕西省教育厅项目(16JK1697);陕西省重点研发计划项目(2017GY-071);陕西省技术创新引导项目(2017XT-005);咸阳市科技计划项目(2017K01-25-3);西安邮电大学研究生创新基金(CXJJLY202003)资助项目。

摘　　要：人耳特征具有良好的唯一性与稳定性等特点,近年来被广泛应用于身份识别领域。针对人耳采集易受头发、耳饰等物品遮挡问题,本文提出了一种基于ERNet的人耳识别方法。该方法在IResNet网络的基础上,引入改进的SE模块,通过融合最大池化与均值池化的统计特性,增强身份相关特征的表示,抑制非相关特征的影响,以此解决在非受控环境下由于遮挡原因造成的识别困难问题。大量实验结果表明,相比较于原网络,改进后的方法识别性能提高较为明显。在同等遮挡条件下,本文所提出的模型具有较好的鲁棒性能。Human ear features were widely used in the field of identity recognition in recent years owing to their good uniqueness and stability.However,the problem that human ear acquisition is easily occluded by hair,earrings,and other objects limits the practical application of human ear biometrics.Here,an ear recognition method based on ERNet is proposed by introducing an improved SE module into the IResNet network.Though integrating the statistical properties of max pooling and mean pooling,this method enhances the representation of identity-related features,suppresses the influence of non-correlated features,and solves the problem that human ears are difficult to recognize due to occlusion in uncontrolled environments.Experimental results indicate that the recognition performance of the improved method is significantly improved compared to that of original network.Meanwhile,the proposed model has good robustness under the same occlusion conditions.

关 键 词：图像融合 注意力机制 卷积神经网络(CNN) 深度残差网络 人耳识别 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程] TP391.41[自动化与计算机技术—控制科学与工程]