Metametric learning-based few-sample website fingerprint identification method

The invention discloses a few-sample website fingerprint identification method based on meta-metric learning. The method specifically comprises the following steps: 1, collecting Tor network anonymous traffic, and preprocessing the Tor network anonymous traffic to form a target data set; 2, selecting an auxiliary data set, and dividing meta-scene tasks according to a supervision target; 3, constructing a meta-metric learning network model, for each meta-scene task, a feature extractor is responsible for extracting original features from network tracking, a task adaptive modulation unit is responsible for adjusting a channel weight, generating a discriminative feature pair of the current task, and generating a sub-scene task; the ridge regression module is responsible for representing each position in the query feature map as a weighted sum of support features of a given category. And 4, training and testing the metric learning network model. The method improves the recognition accuracy and robustness of the s