A multi-scale sentiment recognition network based on deep learning

In recent years, mental health has gradually become a hot topic of concern due to the increase in life stress and the epidemic. Mental health problems may affect the individual's living condition, learning and working efficiency in light cases, or produce mental diseases that prevent normal learning or working. In order to detect psychological problems in a more timely manner and achieve the diagnosis of psychological diseases, people actively research intelligent methods. However, the existing models lead to inaccurate recognition and unsatisfactory accuracy in practical applications due to the angle and occlusion of the face. Therefore, in this paper, we propose a multi-scale deep learning network model for faces, which reduces the effects produced by angles and occlusions by fusing feature information from different scales; and a loss function is proposed to increase the importance of difficult samples in the training process, which makes the model training better. Experimental results on the challenging RAF_DB dataset show that the proposed model exhibits better facial expression recognition accuracy than the existing techniques.