Global Dependency Modeling Target Identification Method Based on Time-Frequency Feature Constraints

doi:10.1109/TGRS.2026.3677296

Global Dependency Modeling Target Identification Method Based on Time-Frequency Feature Constraints

2026-03-25

0

PRE

AI

DOI：10.1109/TGRS.2026.3677296

Original

Original request for help

Save

Abstract

En 中文

In complex interference environments, vibration sensing systems are crucial for monitoring microvibration anomalies, but, in the case of noise interference and partial data loss, the target features will be overwhelmed, making it difficult for data-driven deep learning models to maintain stable recognition performance under multiscene conditions. To solve this problem, we propose a time-frequency constrained global dependency (TFGD) model. The model consists of three parts. First, feature extraction is performed by a convolutional neural network (CNN) and long short-term memory (LSTM) module. Second, a Transformer coding module based on a self-attention mechanism is introduced to establish global dependencies in feature sequences. Most importantly, to solve the key challenges of robust classification due to the lack of low-frequency information and sample class imbalance in noise and vibration data, the logarithmic power spectrum and phase spread spectrum are embedded as physical prior information loss functions. Finally, experimental evaluation of three different datasets, including noise interference scenarios and sample imbalances, showed that the recognition accuracy exceeded 95%, and the standard deviation (SD) of the accuracy was only 0.24%, confirming its effectiveness and application potential in complex environments.

Keywords:

Generalization performance

global dependency

moving target

time-frequency feature

transformer

Global Dependency Modeling Target Identification Method Based on Time-Frequency Feature Constraints

Abstract

Journal

Organization