Weak supervision time sequence behavior positioning method based on adversarial learning

The invention relates to a weak supervision time sequence behavior positioning method based on adversarial learning, and the method comprises the following steps: selecting non-cut video data from a public data set, decomposing each non-cut video into non-repeated frame segments, and then extracting the original features of each frame segment; calculating input feature data Xt of the time sequence continuity branch by using the total number X of the original features; calculating class activation sequence scores and class time sequence attention scores of the basic branch and the time sequence continuity branch; carrying out consistency constraint on the class time sequence attention score and the class time sequence attention score of the overall model; calculating a total loss function of the TEN network model, training the TEN network model at the same time, and then obtaining a trained TEN network model; and inputting to-be-predicted non-clipped video data into the trained TEN network model to obtain beha