Facial Expression Recognition Based on Convolutional Neural Networks and Edge Computing

With edge devices playing an increasingly important role in our daily lives, edge computing and human-computer interaction, especially facial expression recognition, become research central issues in academia and industry. However, surprisingly, utilizing edge computing and neural networks for facial expression recognition has been neglected for many years, very few research can be found. To be focusing on such topics. In this paper, we improve Visual Geometry Group 19 with the idea of residual learning. To be specific, for each block of Visual Geometry Group 19, we add its input to its output. The result of the addition will be the input of the next block. Then, we minimize the size of our model by pruning and post-training quantization to achieve a higher efficiency and maintain the model's accuracy at the same time when deploying it on edge devices. The experiment result shows that our model has a 98.99% accuracy on the CK+ dataset. Besides, when deploying on edge devices, its inference time is less than many other popular neural networks that are designed for deploying on edge-devices.