A cross-scenario underwater acoustic target classification method based on model transfer learning

Through the model transfer learning method, the convolutional neural network and knowledge distillation technology are used to solve the problem of weak generalization ability in cross-scene applications in the existing technology, and the classification of water acoustic targets with high generalization ability and accuracy is achieved. Strongly capable and suitable for different application scenarios.

CN113780242BActive Publication Date: 2025-02-07THE 54TH RESEARCH INSTITUTE OF CHINA ELECTRONICS TECHNOLOGY GROUP CORPORATION
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202111160013.1
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2021-09-30
Publication Date
2025-02-07
Estimated Expiration
2041-09-30

AI Technical Summary

Technical Problem

When the existing technology is applied across scenarios, the generalization ability is weak, the classification model needs to be retrained, and the data acquisition and labeling cycle is long, which is expensive, making it difficult to quickly adapt to different application scenarios.

Method used

Using a cross-scene hydroacoustic target classification method based on model transfer learning, filtering and denoising and time spectrum extraction of the water acoustic target signals received by sonar devices, using convolutional neural networks for training, and introducing knowledge distillation technology to extract The target type soft label forms a classification model in new scenarios.

Benefits of technology

It achieves high generalization ability, high accuracy and good robustness for cross-scenario applications, and can quickly adapt to new environments, new goals or new tasks. It only requires a small number of target domain samples to achieve good classification performance.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN113780242B_ABST
    Figure CN113780242B_ABST
Patent Text Reader

Abstract

The present invention discloses a cross-scenario underwater acoustic target classification method based on model transfer learning, which belongs to the technical field of passive reconnaissance of underwater acoustic targets and is suitable for classifying underwater acoustic target signals received in array form. This method uses the target signal features extracted by the time-frequency analysis method as shared feature space samples for cross-scenario classification, trains the convolutional neural network, and first encodes the knowledge in the model parameters, model prior knowledge and model architecture in the source data domain to form a stable classification capability. Then it is migrated to the target scene, and the category relationship knowledge in the model is fitted in the new sample and label space based on the knowledge distillation technology to realize the classification capability. The present invention solves the problem of poor generalization ability of existing underwater acoustic target classification methods in cross-scenario applications by using model transfer learning technology, and can quickly adapt to new environments and new tasks under the conditions of small samples and data imbalance, greatly saving the time, energy and resources consumed by large-scale data collection and model retraining in the target domain.
Need to check novelty before this filing date? Find Prior Art

Citation Information

Patent Citations

  • CNN and transfer learning based disease intelligent identification method and system

    AU2020103613A4

  • A general object detection framework based on picture distillation and its implementation method

    CN109344897A