Adv-FVMamba: Anxiety Disorders Recognition in Imbalanced Video Datasets Using Adversarial Entropy Loss

There is a notable scarcity of effective datasets tailored to Chinese adolescents within the realm of anxiety disorder research, further compounded by the prevalent issue of imbalanced data distributions in existing datasets. Therefore, this study serves as the basis for solving the above problems by establishing a dataset of anxiety disorders among Chinese adolescents, derived from responses to the Mental Health Testing (MHT) scale targeted at primary and secondary school students. High-definition cameras capture videos of 145 adolescents' faces during the answering process. After evaluation, data from 112 adolescents, aged 9–11 years old, 55 boys and 57 girls, are deemed valid. And we propose Adv-FVMamba, which is designed for anxiety disorders recognition in imbalanced video datasets using adversarial entropy loss. We compare our model to other models in the dataset we established, and the recognition performance of ours is higher than others, with an F1-score of 0.75123. Experimental results verify the effectiveness of our proposed method in anxiety disorders recognition.