摘要
LeNet-5卷积神经网络在手写数字库上取得了很好地识别效果,但在表情识别中识别率很低.改进了LeNet-5卷积神经网络,使用浅层卷积结构,连续经过1×1和3×3的卷积层,在每一层的卷积后,加上Z-score标准化处理,使用性能更好的Relu激活函数,此函数计算速度快,减少梯度弥散问题;输出层用softmax函数,该层输出表情图像的概率.仿真结果表明,在JAFFE表情数据库上,即使在小样本数据集的情况下,算法识别率达到79.81%,识别单幅人脸表情图像的平均耗时为0.353 s.
The LeNet-5 convolutional neural network achieves good recognition on handwritten digit libraries,but has a low recognition rate in expression recognition.LeNet-5 convolution neural network is improved by using shallow convolution structure and passing through 1*1 and 3*3 convolution layers successively.After convolution of each layer,plus Z-score standardization,Relu activation function with better performance is used because it calculates quickly and reduces gradient dispersion problems.Softmax function is used in output layer to output probability of expression image.The simulation results show that the recognition rate of the algorithm reaches 79.81% even with small sample data sets on JAFFE expression database,and the average time to recognize a single face expression image is 0.353 s.
作者
赵彩敏
刘国红
ZHAO Caimin;LIU Guohong(School of Electrical&Mechano-Electronic Engineering,Xuchang University,Xuchang 461000,China)
出处
《许昌学院学报》
CAS
2021年第2期113-116,共4页
Journal of Xuchang University
基金
许昌学院校级科研项目(2020YB011)。
关键词
表情识别
卷积神经网络
激活函数
Z-score标准化处理
facial expression recognition
convolutional neural network
activation function
Z-score standardized processing
