2025_programme: Transformer-CNN-Based Algorithm for Underwater Maneuvering Target Tracking

Back


  • Day: June 19, Thursday
      Location / Time: TBA at 15:30 - 16:30
  • Last minutes changes: -
  • Session: Poster session
    Organiser(s): N/A
    Chairperson(s): N/A
  • Lecture: Transformer-CNN-Based Algorithm for Underwater Maneuvering Target Tracking
    Paper ID: 2194
    Author(s): Tianhang Ji, Xiaochuan Ma, Yu Liu, Xiaomei Wang, Chao Feng, Pengzhuo Li, Yubo Hu
    Presenter: Tianhang
    Abstract: Due to the extended length of time series and the maneuverability of target motion, the performance of underwater maneuvering target tracking using Recurrent Neural Networks (RNN) and Long Short-Term Memory networks (LSTM) degrades. To address this issue, we propose a maneuvering target tracking algorithm that combines Transformer network structure with Convolutional Neural Network (CNN). This algorithm accurately estimates the true motion state of the target by introducing a novel residual mapping between the observed trajectory and the true trajectory. The algorithm consists of three main modules: an input module, an encoder, and a decoder. The input module encodes the state features of the observed trajectory, extracting the correlation information between states. The encoder, based on the attention mechanism, extracts features from the target's historical trajectory, enhancing the ability to capture the target's maneuvering behavior. The decoder utilizes a convolutional neural network to output the residual between the observed trajectory and the true trajectory, thereby improving the tracking capability of maneuvering targets. Additionally, we propose a new loss function that optimizes weight allocation for the magnitude differences between velocity and position, enhancing the model's accuracy in estimating target velocity. Monte Carlo simulation experiments demonstrate that the proposed algorithm outperforms RNN and LSTM in terms of both position accuracy and speed accuracy.
  • Corresponding author: Prof Xiaochuan Ma
    Affiliation: Institute of Acoustics,Chinese Academy of Sciences
    Country: China

Back