基于全局平均池化的轻量级表情识别模型及其在移动端的应用，首发论文

Lightweight Facial Expression Recognition Model and Its Application on Mobile Based on Global Average Pooling

张东晓（1980-），男，副教授，硕导，主要研究方向：视频与图像处理、机器学习

Abstract：A shallow convolution neural network is constructed for the expression recognition of mobile terminal. Three groups of stacked convolution layers help to enlarge the receptive field and facilitate feature extraction; The global average pooling layer, which substitutes for additional full connection layer, greatly reduces the amount of parameters. After training model in the FER-2013, the accuracy rate on the test set can reach 0.7; After freezing the other layers except the average pooling, Fine-tuning on the CK + data set make the accuracy rate on the test set reach 0.96. The model is transformed into the core ML model, and a real-time App about expression recognition is established on the iOS with Xcode platform. It can run stably and smoothly on the iPhone 8 plus, and the effect reaches the expectation.

Keywords： Artificial Intelligence Facial Expression Recognition CNN Global Average Pooling Google Colab Core ML