改进多尺度卷积神经网络的人脸表情识别研究

为了有效改善现有人脸表情识别模型中存在的信息丢失严重、组件间相对空间联系不密切的问题,提出了一种改进的多尺度卷积神经网络模型,通过构建深层多尺度卷积神经网络,使模型能够挖掘出更多潜在的特征信息;通过特征融合促进信息的流通和重利用,减少池化操作所引起的重要信息丢失,使得模型具有更好的学习能力;通过控制每层多尺度卷积神经网...

Research on facial expression recognition based on improved multi-scale convolutional neural networks

In order to effectively improve the problems of severe information loss and poor relative spatial connection between components in existing facial expression recognition models, an improved multi-scale Convolutional Neural Network (Multi-scale CNN) model is proposed. By constructing deep multi-scale convolutional neural networks, the model is able to uncover more information about potential features. Through feature fusion, we can promote the circulation and reuse of information, reduce the loss of important information caused by pooling operation, and make the model have better learning ability. By controlling the convolution kernel size of each layer of multi-scale convolution neural network to balance the relationship between global features and local features, this model enhances the relative spatial relationship between different components and avoid the redundancy of feature map channel information. The verification of two different facial expression recognition datasets JAFFE and FER-2013 shows that the accuracy of the algorithm on test sets reaches 95.45% and 76.56% respectively, which proves the effectiveness and the advanced nature of the algorithm.