融合注意力机制与残差网络的人耳识别方法

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。

Ear recognition method combining attention mechanism and residual network

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.