Bidirectional-Deep-Bayesian-Smoother

This is the content of the code in our paper "Bidirectional Joint State-Memory Deep Bayesian Smoother".

Requirements

The code contains the tensorflow implementation of EGBRNN. Different experiments are placed in different paths, each with a corresponding requirement.txt file. To install requirements:

pip install -r requirements.txt

Data

You can also regenerate or download the raw data.

For "Aircraft tracking", you can download raw flight records with format "lt6" from the open resource repository. The matlab and python programs used to process the raw data are in the following paths:

├── Raw_data_processing
   ├── lt6_to_mat  #  Step 1. Matlab code for processing raw flight records in lt6 format.
   └── mat_to_npy  #  Step 2. Python code for processing data in mat format.

For "Vehicle localization", you can download the dataset by following the references provided in our paper.

In addition, we also provide examples of the preprocessed data.

Training & Testing

We have provided annotations for the code related to aircraft trajectory smoothing and vehicle localization, so you can easily locate and run them based on the directory structure.

Note

If your work involves this code, please pay attention to our our papers:

[1]Yan S, Liang Y, Zheng L, et al. Explainable Gated Bayesian Recurrent Neural Network for Non-Markov State Estimation[J]. IEEE Transactions on Signal Processing, 2024. (code : https://github.com/DeepBayesEst/EGBRNN_TSP)

[2]Yan S, Liang Y, Zhang H, et al. Explainable Bayesian Recurrent Neural Smoother to Capture Global State Evolutionary Correlations[J]. arXiv preprint arXiv:2406.11163, 2024.

[3]Yan S, Liang Y, Zheng L, et al. Memory-biomimetic deep Bayesian filtering[J]. Information Fusion, 2024, 112: 102580.