Timing data visualization: tactical intent recognition and portable framework

Abstract

By transforming time series into images, a robust and transferable tactical intent recognition framework was proposed, which integrated curve filtering technology and the EfficientNetV2 image recognition network. Curve filtering technology effectively reduced redundancy in numerous time-domain features, model parameters, and training time, an enhanced Gramian angular field (GAF) method was proposed to encode time series into images, enhancing the feature extraction capabilities of convolutional neural networks. The EfficientNetV2 network was adept at processing intent images and could serve as a pre-trained model, facilitating transfer learning across different systems. Experimental results demonstrate that the proposed framework achieves over 0.99% higher accuracy compared to machine learning and deep learning methods, exhibiting superior performance, scalability, robustness, and transferability.

Keywords

• time series coding;intention recognition;image classification;curve filtering;Gramian angular field;EfficientNetV2