基于多特征融合卷积神经网络的年龄预测  

Age Prediction Based on ulti-feature Fusion Convolutional Neural Network

作　　者：廖黄炜 马燕[1] 黄慧 LIAO Huang-wei;MA Yan;HUANG Hui(School of Information,Mechanical and Electrical Engineering,Shanghai Normal University,Shanghai 200234,China)

机构地区：[1]上海师范大学信息与机电工程学院,上海200234

出　　处：《计算机技术与发展》2022年第10期58-64,共7页Computer Technology and Development

基　　金：国家自然科学基金项目(61373004)。

摘　　要：使用卷积神经网络预测人脸图像年龄。目前标注的表观年龄公开数据集不足且人数规模小,小样本训练紧凑模型成为年龄预测的研究方向之一。为提高卷积神经网络对图像中年龄特征提取能力和小样本数据迁移能力,并减少对大型图像分类数据集的依赖,提出多特征融合卷积神经网络结构的年龄预测方法。该方法设计多特征融合模块融合两者特征向量获取图像的多层次信息;主干特征提取卷积网络中引入注意力机制模块,增强年龄特征提取能力;设计序列化年龄标签的组合损失函数,使得年龄特征充分拟合年龄标签。在IMDB-WIKI预训练下,该方法在MORPH上取得的平均误差为1.951,ChaLearn15上平均误差为3.128。实验结果表明,该方法的训练和迁移成本低,与许多最新方法相比取得了较高的预测精度。Convolutional neural network is used to predict the age of face images.At present,the publicly labeled apparent age dataset is insufficient.Small sample training compact model has become one of the research directions of age prediction.In order to improve the ability of convolutional neural network to extract age features in images and the ability of small sample data migration,and reduce the dependence on large image classification datasets,a multi-feature fusion convolutional neural network structure age prediction method is proposed.This method designs a multi-feature fusion module to fuse the feature vectors of the two to obtain the multi-level information of the image.The main feature extraction convolutional network introduces the attention mechanism module to enhance the ability of age feature extraction.A combined loss function of serialized age labels is designed to make age features fully fitting the age label.Under the pre-training of IMDB-WIKI,the average error obtained by the proposed method on MORPH is 1.951,and the average error on ChaLearn15 is 3.128.The experimental results show that the training and migration cost of the proposed method is low,which has achieved higher prediction accuracy compared with many state-of-the-art methods.

关 键 词：特征融合 年龄预测 卷积神经网络 深度学习 小波变换 GABOR 

分 类 号：TP753[自动化与计算机技术—检测技术与自动化装置]