Facial Expression Recognition With Deep Learning Approach For Detecting Stress

A workplace stress detection system using face images require a low-cost installation and does not interfere with the movements of workers, thus the system will be the better promising approach than physiological data based approach using wearable devices. As a preliminary research, we proposed a methodology for emotion recognition to detect workplace stress. We trained various structures of convolutional neural networks (CNNs) with real-world affective image dataset (RAF-DB). Among the trained CNNs, VGG16 showed the highest accuracy on test set. In order to further improve the classification performance, we used generative adversarial network (GAN) as a data augmentation method. After post-processing generated images by the GAN, we fine-tuned VGG16 with mixed dataset, which included original training images and post-processed generated images, and then we found that the classification performance was slightly improved. In conclusion, The developed network showed 78.5 % accuracy for face emotion and is expected to be used in algorithms for stress detection in the future.1