关键词: 卷积神经网络, 局部二值模式, 特征融合, 双通道模型, 表情识别, 图像信息熵, 婴幼儿, 支持向量机

Abstract:

In view of the problems of low rate of infant expression recognition and the insufficient extraction of complex features, an infant expression recognition algorithm based on MobileNetV2 and LBP dual-channel feature fusion is proposed. The first channel uses the improved MobileNetV2 network to quickly and accurately extract the global features of facial expressions. The second channel divides the original input image into blocks, and uses image information entropy to construct weights, and extracts block-weighted LBP histogram features to highlight the regions with rich expression information. The output vector of the dual-channel model is fused to improve the feature expression ability, and the support vector machine is used to replace the Softmax layer for expression classification. Experiments show that the use of fusion features has a better classification effect than a single feature, and in the self-built infant expression data set, the accuracy of expression recognition can reach 85.71%.

Key words: convolutional neural network, local binary pattern, feature fusion, dual-channel model, expression recognition, image information entropy, infant, support vector machine