Lightweight vit depression facial image detection with integrated attention mechanism

In order to solve the problems of limited ability of extracting local information and excessive parameters in the diagnosis model of major depression (MDD), an image depression recognition algorithm based on the lightweight CNN+Transformer model of transfer learning (attn_ConvNeXt _MobileVit) was proposed to evaluate depression from facial images. The model adopts CASIA WebFace pre training model to improve the generalization ability of the model on small data sets; CA attention mechanism and ConvNeXt convolution module optimized by SimAm are used to enhance the extraction of local information; The lightweight mobile ViT module combined with global local information is used to maintain a low parameter amount while improving the depression recognition ability of the model. The experimental analysis of the public dataset AVEC2014 shows that compared with other methods, the predicted results of depression scores of subjects are better, with RMSE of 8.37 and MAE of 6.28, which verifies the effectiveness of the model and algorithm.