A liquefied gas leakage video image intelligent monitoring method based on knowledge distillation

By employing a knowledge distillation-based approach, combined with selective attention mechanisms and feature reconstruction algorithms, the liquefied gas leak monitoring model was optimized. This addressed the issues of poor positioning quality caused by fuzzy liquefied gas boundaries and high monitoring accuracy in resource-constrained environments, achieving efficient and accurate liquefied gas leak monitoring.

CN122115815APending Publication Date: 2026-05-29PETROCHINA CO LTD +1
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202610189383.4
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Priority Date
2025-09-05
Filing Date
2026-02-10
Publication Date
2026-05-29

AI Technical Summary

Technical Problem

Existing liquefied gas leak monitoring technologies suffer from problems such as poor positioning quality due to fuzzy liquefied gas boundaries in the model, poor real-time performance, lack of intelligent discrimination capabilities, and difficulty in balancing monitoring accuracy and efficiency in resource-constrained environments.

Method used

We employ a knowledge distillation-based approach, utilizing selective attention, extremum removal, numerical stability, contrastive learning, and feature similarity techniques. We design local masking, feature reconstruction, and regional sample contrastive distillation algorithms to construct a student model. The output of the teacher model guides the training of the student model, optimizing the network structure and improving feature representation and multi-scale feature detection capabilities.

Benefits of technology

It improves the accuracy and resource performance of liquefied gas leak monitoring, solves the problem of poor positioning quality caused by the fuzziness of liquefied gas boundaries in the model, realizes efficient monitoring in resource-constrained environments, reduces misjudgments and false alarms, and improves monitoring accuracy and robustness.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122115815A_ABST
    Figure CN122115815A_ABST
Patent Text Reader

Abstract

The present application relates to the field of power distribution network voltage control, and particularly relates to a liquefied gas leakage video image intelligent monitoring method based on knowledge distillation, which comprises the following steps: acquiring a data set of liquefied gas leakage images, and dividing the data set into a training set, a validation set and a test set according to a proportion; using an open-source public data set to pre-train a knowledge distillation model to obtain a teacher model, extracting the soft label of the teacher model and constructing a student model, the student model being a liquefied gas leakage monitoring model based on knowledge distillation; setting preset training conditions for the teacher model and the student model; based on the preset training conditions, training the teacher model and the student model using the training set, the teacher model guiding the training of the student model through a knowledge distillation method to obtain a target student model, so that the target student model can approach the performance of the teacher model; and monitoring the liquefied gas through the target student model to obtain a liquefied gas leakage monitoring result.
Need to check novelty before this filing date? Find Prior Art