Face video-based depression automatic identification method

The invention relates to an automatic depression identification method based on a face video, and the method comprises the following steps: obtaining to-be-identified face video data, and carrying out the preprocessing of the to-be-identified face video data, and obtaining a face video sub-sample Vc; constructing a deep learning model LDSML for joint learning of depression label distribution and label measurement; inputting the subsample Vc into an LDSML deep learning model for training; using a Softmax-T activation function with a temperature coefficient at the output end of the model to obtain predicted depression label distribution p, and calculating a mathematical expected value E of the label distribution; in the prediction stage, the mathematical expected value E of the predicted depression label distribution P is used for representing a final depression prediction result. On the premise that a training data set is not increased, the feature capture capability of the deep learning model on the facial de