基于级联网络的年龄性别识别研究

Age and gender recognition based on Network Cascade

提出了一种基于深度学习技术的年龄和性别识别方法,建立了人脸识别硬件与软件系统。运用经典的反向传播算法确定预测函数的权重矩阵和偏差。针对单个识别网络准确率不高的问题,采用多个神经网络级联的方式,对输入的目标特征进行多次判定。通过设计一种投票竞争算法,让级联网络的识别结果自动进行竞争,获胜者作为最终的预测结果。预测结果与实验数据对比表明采用级联网络可有效提高对年龄性别的识别准确率,级联后的识别准确率分别达到了88%和82.61%。

An age and gender recognition method based on deep learning technology was proposed, and a face recognition hardware and software system was established. The classical backpropagation algorithm is used to determine the weight matrix and bias of the prediction function. For the problem that the accuracy of a single identification network is not high, multiple neural network cascades are used to determine the input target features multiple times. By designing a voting competition algorithm, the recognition results of the cascaded network are automatically competed, and the winner is the final prediction result. The comparison between the prediction results and the experimental data shows that the cascading network can effectively improve the recognition accuracy of age and gender, and the recognition accuracy after cascading reaches 88% and 82.61% respectively.