In the field of psychological crisis intervention and prevention, the application of deep learning algorithm provides a new perspective for the extraction and analysis of psychological crisis features. As mental health issues become increasingly prominent, existing research still faces challenges in terms of the accuracy of feature extraction and the efficiency of data analysis. Based on deep learning algorithm, this paper extracts and analyzes psychological crisis features and constructs a psychological state feature recognition model. By studying deep learning algorithm, convolutional neural network and long short-term memory network are used to identify psychological crisis features, and the effectiveness of the model is verified by experiments. The experimental results show that this model performs well in terms of psychological feature extraction accuracy and data analysis efficiency, with the highest extraction accuracy of 100%, which provides strong support for the early identification of psychological crisis.