Semi-supervised learning video behavior identification method based on Soft-Teach

The invention discloses a semi-supervised learning video behavior recognition method based on Soft-Teach, and relates to the technical field of video processing. Obtaining video sample data, and dividing the video sample data into labeled data and unlabeled data; and inputting the labeled data and the unlabeled data into the trained video behavior recognition model to obtain a classification result of a final video behavior category. The method depends on a small amount of annotation data and a large amount of annotation data for training, so that the workload of a video data annotation task is reduced; a local feature learning module is introduced for the problem of offset of a lost space of Transform, and a sequence pooling technology is introduced to optimize a predicted value in order to better utilize data of a whole sequence. 本发明公开了基于Soft-Teacher的半监督学习视频行为识别方法，涉及视频处理技术领域。获取视频样本数据，将视频样本数据划分为有标签数据和无标签数据；将有标签数据和无标签数据输入至训练好的视频行为识别模型中，得到最终视频行为类别的分类结果。本发明依赖少量标注数据以及大部分标注数据进行训练的方法，降低了视频数据标注任务的工作量；针对Transformer的