摘要： 双相抑郁症(Bipolar Disorder)会使人们因为严重的情绪问题无法参与正常的社会生活，甚至导致自残和自杀行为。为了准确检测患者当下心理状态以协助医生进行更精准的治疗，提出了一种结合注意力机制的CNN-LSTM网络的混合模型的双相抑郁症检测方法。该方法首先使用在人脸表情数据集上微调的Resnet50模型提取视频帧的空间特征，其次通过结合注意力机制的LSTM网络提取帧之间的时序信息去检测双相抑郁症。在AVEC2018双相抑郁症数据库开发集上，验证了该方法的有效性。

中图分类号： TP18
文献标识码： A
DOI： 10.19358/j.issn.2096-5133.2022.05.011
引用格式： 穆家宝. 结合注意力机制的CNN-LSTM的视频中双相抑郁症检测方法[J].信息技术与网络安全，2022，41(5)：72-76.

Bipolar disorder detection in videos by integrating attention mechanism based on CNN-LSTM

Abstract： Bipolar disorder can prevent people from participating in normal social life because of severe emotional problems, and even lead to self-harm and suicidal behavior. In order to accurately detect the patient′s current psychological state to assist doctors in more accurate treatment，in this paper, a hybrid model of CNN-LSTM network combined with attention mechanism is proposed for bipolar depression detection. The method firstly uses the Resnet50 model fine-tuned on the facial expression dataset to extract the spatial features of video frames, and secondly uses the LSTM network combined with the attention mechanism to extract the temporal information between frames to detect bipolar depression. The effectiveness of this method was verified on the development set of the AVEC2018 bipolar depression database.

Key words : bipolar disorder detection；convolutional neural network；long short-term memory