Emotion Recognition Method for Community Teenagers Based on RepNanoNet

To accurately and efficiently identify the emotional state of community teenagers, a lightweight emotion recognition method based on RepNanoNet was proposed to better support the mental health development of teenagers. RepNanoNet is based on MobileNetV2 and introduces various improvements to improve the emotion recognition ability and reasoning efficiency of the model. Rep-Depthwise Conv enhances the feature extraction ability of the model by combining the Structural Re-parameterization method, while Channel Shift promotes information interaction between channels to optimize feature distribution. Rep-Inverted block combines the advantages of two lightweight blocks to further improve the learning ability and computational efficiency of the model. Experimental results show that RepNanoNet achieves an accuracy of 58.28% on the FER-2013 dataset, while maintaining a low number of parameters and FLOPs, which are 2.60M and 0.42G respectively.