Time series data classification method and device based on content awareness embedding, equipment and medium

By preprocessing and segmenting human motion time series data, and using a pre-trained model to extract deep semantic features and calculate similarity, the problems of lack of content awareness in the embedding layer and inability of the output layer to model category distribution in time series classification are solved, thereby improving classification accuracy and performance.

CN121580261AActive Publication Date: 2026-02-27NAT UNIV OF DEFENSE TECH
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202610106667.2
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-01-27
Publication Date
2026-02-27
Estimated Expiration
2046-01-27

AI Technical Summary

Technical Problem

Existing time series classification methods suffer from poor classification performance due to the lack of content awareness in the embedding layer and the inability of the output layer to effectively model category distribution.

Method used

By collecting time-series data during human movement, preprocessing and segmenting are performed, content-aware embedding sequences are calculated, deep semantic feature vectors are extracted using a pre-trained model, and these vectors are projected onto a unit hypersphere to calculate similarity. Classification is then performed by combining cross-entropy and orthogonal regularization loss functions.

Benefits of technology

It improves the accuracy and performance of time series data classification, effectively identifies human movements, and solves the problems of lack of content awareness in the embedding layer and inability of the output layer to model category distribution.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN121580261A_ABST
    Figure CN121580261A_ABST
Patent Text Reader

Abstract

The invention discloses a time series data classification method and device based on content awareness embedding, equipment and a medium, and relates to the technical field of neural networks, and the method comprises the steps: preprocessing time series data of a human body in a motion process; segmenting the preprocessed time sequence data to obtain time sequence blocks; calculating a content awareness embedding sequence, determining a sequence block position embedding sequence corresponding to the time sequence block, and generating an embedding sequence; extracting the embedded sequence by using a pre-training model to obtain a feature vector; performing normalization processing on the feature vector and the prototype vector, projecting the normalized feature to a unit hypersphere, calculating the similarity between the projected normalized feature and the prototype vector, and calculating a time sequence data classification prediction result; linear layer classification is carried out on the time sequence data, corresponding actions of a human body during movement are identified, and the problem that classification performance is poor due to the fact that an embedded layer lacks content perception ability and an output layer cannot effectively model category distribution is solved.
Need to check novelty before this filing date? Find Prior Art

Citation Information

Patent Citations

  • Time series data processing method and device, equipment and medium

    CN120578888A

  • Old people emotion recognition method and device based on multi-modal perception

    CN120899250A