Attention Mechanism-ResNet-based Feature Recognition of Depressed Patients

In this paper, a multimodal feature fusion model for the face and eyes based on convolutional neural networks is proposed, which excels in diagnostic accuracy. The key features of the face and eyes are extracted by watching videos with emotional features and playing an eye movement mini-game. The most important features were effectively extracted by increasing the number of features using convolutional layers. The model has a high accuracy of 87.6% in recognizing depressed patients. In this paper, we hope to extend the generalization of the model and improve the recognition of patients with mild depression, in order to help more patients to be diagnosed and treated at an early stage. This paper realizes two innovations in modeling method and diagnostic purpose. In the modeling method, this paper chooses ResNet convolutional neural network,adds channel and spatial attention mechanism, uses fuzzy synthesis algorithm to couple features of face and eyes, and finally classifies and recognizes by A-SoftMax. The facial and eye images of the examined patients are used as input features, feature extraction and classification of depression diagnosis type based on convolutional neural network. For diagnostic purposes, this more focuses on the recognition rate of patients with mild depression in the hope that depressed patients can get timely and effective treatment at the initial stage.