摘要
为了进一步提高年龄估计的精度,提出一种基于深度学习与有向无环图SVM的局部调整年龄估计算法。在训练阶段,将经过VGGFace2数据集预训练的SE-ResNet-50网络进行微调,待到收敛时提取出全连接层,将其首尾相连形成的向量作为表征并训练出多个One-Versus-One SVM。在测试阶段,将待估计人脸图像送入SE-ResNet-50以得到一个较为粗略的年龄估计值;设定具体邻域;将训练而成的SVM组合为一个有向无环图SVM并以全局估计值为中心进行精准的年龄估计。为了表明算法的普适性,在不同种族的MORPH和AFAD图像集中进行实验,结果验证了算法的有效性。
In order to further improve the accuracy of age estimation, a local adjusted age estimation algorithm is proposed based on deep learning and directed acyclic graph SVM. In the training phase, the SE-Resnet-50 network pre-trained by VGGFace2 data set was fine-tuned, until convergence to extract the whole connection layer, and the vector formed by connecting them end to end was used as a representation and multiple One-Versus-One SVM were trained. In the testing stage, the face images to be estimated was sent to SE-ResNet-50 to get a rough age estimate, the specific neighborhood was set, the trained SVM was combined into a directed acyclic graph SVM, and the global estimate value was taken as the center for accurate age estimation. In order to show the universality of the algorithm, experiments were carried out on different races of MORPH and AFAD image sets, and the results verified the effectiveness of this algorithm.
作者
赵卫
刘渊
Zhao Wei;Liu Yuan(Wuxi Commercial Vocational College,Wuxi 214153,Jiangsu,China;Jiangnan University,Wuxi 214122,Jiangsu,China)
出处
《计算机应用与软件》
北大核心
2023年第1期189-195,共7页
Computer Applications and Software
基金
国家自然科学基金项目(61972182)。
关键词
年龄估计
深度学习
有向无环图SVM
局部调整
Age estimation
Deep learning
Directed acyclic graph SVM
Local adjustment
