基于深度学习算法的小样本人耳识别  

Small Sample Ear Recognition Based on Deep Learning

作　　者：李万相 田莹 LI Wan- xiang;TIAN Ying(Northeastern University,Liaoning Shenyang 110819,China;University of Science ＆ Technology,Liaoning Anshan 114051,China)

机构地区：[1]东北大学,辽宁沈阳110819 [2]辽宁科技大学,辽宁鞍山114051

出　　处：《计算机仿真》2018年第8期246-250,369,共6页Computer Simulation

基　　金：国家自然基金(72472081);辽宁省教育厅资助项目(L2014115)

摘　　要：虽然人耳识别取得很大进展,但在非控制条件下人耳识别容易受姿态变化和遮挡等因素的影响,仍存在很多问题;此外人耳识别实用系统往往难以获得待识别个体的多个样本。针对上述两点提出新的基于深度学习的小样本人耳识别方法。首先把人耳识别主要模块构建成一个卷积神经网络模型,并在其全连接层后面加一个K-近邻层以均衡所有分类,解决卷积神经网络对于小样本的过拟合问题。使用构建的模型自动学习人耳图像的特征,最后利用支持向量机对提取的人耳特征进行分类识别。实验结果与后续分析证明,与传统的机器学习识别方法相比,改进的深度学习模型具有更高的人耳识别率和更强的鲁棒性。Human ear recognition has made great progress, but in non - controlled conditions ear recognition is easy to be influenced by posture and occlusion, and many problems still exist. Also, it is difficult to obtain a large number of ear samples of each people. To solve the problems above, a new small sample ear recognition algorithm based on deep learning is proposed. At first the main module of ear recognition was constructed into a convolutional neural network structure. After full connection, a K - nearest neighbor layer was added to balance all classification, and to solve the overfitting problem of convolutional neural network based on small sample. Then the ear image fea- ture was learnt by constructed depth learning network automatically. Finally, the support vector machine was used to classify the extracted features. The experimental results and analysis show that compared with the traditional machine learning algorithm, the improved depth learning model has higher recognition rate and stronger robustness.

关 键 词：深度学习 卷积神经网络 支持向量机 人耳识别 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]