FelixOtt/gnss_pseudo_labeling

Achieving Generalization in Orchestrating GNSS Interference Monitoring Stations Through Pseudo-Labeling

Latest update: 2024-10-30

Project size: 13 GB

Introduction

We propose an machine learning approach that achieves high generalization in classifying interference through orchestrated monitoring stations deployed along highways. We present a semi-supervised approach coupled with an uncertainty-baased voting mechanism by combining Monte-Carlo dropout and Deep Ensembles that effectively minimizes the requirement for labeled training samples to less than 5% of the dataset while improving adaptability across varying environments. Our method demonstrates strong performance when adapted from indoor environments to real-world scenarios.

Datasets

The datasets con be downloaded from the following websites:

For our spectrum highway dataset 1 recorded along a highway bridge, see the following website: https://gitlab.cc-asp.fraunhofer.de/darcy_gnss/FIOT_highway

For our spectrum highway dataset 2 recorded along a highway bridge, see the following website: https://gitlab.cc-asp.fraunhofer.de/darcy_gnss/fiot_highway2

For our dataset recorded in a controlled environment with a high-frequency antenna, see the following website: https://gitlab.cc-asp.fraunhofer.de/darcy_gnss/FIOT_LC_laboratory

For our dataset recorded in a controlled environment with a low-frequency antenna, see the following website: https://gitlab.cc-asp.fraunhofer.de/darcy_gnss/controlled_low_frequency

The folder GNSS_dataset contains the Spectrum datasets with the corresponding labels in .txt files for the train-test split. One sample in the text file is defined as for example:

data/157369.npy 0

In this repository, separate .txt files with labels are included such that a combination and cross-validation of all GNSS datasets is possible.

Method

The following figure presents a comprehensive overview of the proposed method. Initially, a ResNet18 model, which has demonstrated high performance in GNSS classification tasks, is pre-trained on a dataset collected in an indoor environment. The objective is to adapt this model to new scenarios, specifically a highway environment, using only a limited number of labeled samples. We evaluate the model using 0.5%, 1%, and 5% of labeled samples from the original dataset. Subsequently, $M$ ResNet18 models, denoted as $m_i$ where $i \in {1, 2, \ldots, M}$, are trained. Experiments are conducted with $M \in {1, 2, 4}$ models. The models then predict the class for the remaining unlabeled dataset using a voting mechanism, with selection criteria based on uncertainty estimates from Monte Carlo and Deep Ensembles. These pseudo-labeled data are subsequently combined with the labeled dataset, and the models are retrained. Finally, the models perform predictions on the test dataset for the interference classification task.

Training and Evaluating the Models

1. Use train_resnet.py to train a model

   --run_num defines the number of the training run

   --data defines the dataset which is used example: 0,1 will train the datsets 1F and 2F together (integers seperated with , In the given trainings this is always 1)

   --augment defines whether to use augmentation or not (integer 0 or 1)

   --multi_runs defines the number of times each sample is evaluated during validation (integer if x > 1 each sample is run x times with a small dropout else each sample is run once without dropout)

   --use_uncertainty_aware_pseudo_labeling (integer if 1 use the pseudo labeling function 5 from https://arxiv.org/pdf/2101.06329 for pseudo label selection)

   --use_neg_learning (integer if 1 use negatives in addition to positives for learning https://arxiv.org/pdf/2101.06329)

2. Use test_resnet.py to test a trained model on all test datasets

   --folder defines the string name of the folder in which the training to be tested is stored (str, example: resnet_pseudo for the folder results_resnet_pseudo)

   --run_num defines the run number of the training to be tested (integer)

3. Use collect_results.py to collect all results from a defined folder

   --folder defines the string name of the folder in which the test results to be colected are (str, example: resnet_pseudo for the folder results_resnet_pseudo) collected results will be stored in the provided folder + "/results/"

   --data defines the tested on dataset to be collected (integer for the provided trainings and tests this is always 1)

   --repeats defines the number of pseudolabeling steps (integer for the provided trainings and tests this is always 6)

4. plot_results_pseudo.py creates overview plots for pseudo labeling

   --folder defines the string name of the folder in which the results are (str, example: "" for the folder results_resnet_pseudo; "uncertainty" for the folder results_resnet_pseudo_uncertainty)

   --data defines the tested on dataset to be collected (integer for the provided trainings and tests this is always 1) images will be stored in the provided folder + "/images/" as is this script will only work for the provided trainings and tests The created overview plots are named as follow: overview_pseudo_dataset(always1)_MonteCarloSampling(if 0 the only 1 repeat per sample, if 1 the 5 repeats per sample)_DeepEnsemble(if 0 only 1 net, if 1 then 2 nets and if 2 then 4 nets)_pretrained(if 0 then pretrained on Highway1, if 1 then pretrained on ControlledLowFreq, if 2 then pretrained on ControlledHighFreq) For uncertainty_param_search the names are: overview_pseudo_dataset(always1)_kappaN(0: 0.05, 1: 0.005, 2: 0.0005)_tauN(0: 0.1, 1: 0.05, 2: 0.01)

References

For more information of the dataset and results, see our publication. If you use our datasets for your research, please consider citing:

@inproceedings{heublein_feigl_posnav,
  author = {Lucas Heublein and Tobias Feigl and Alexander Rügamer and Felix Ott},
  title = {{Achieving Generalization in Orchestrating GNSS Interference Monitoring Stations Through Pseudo-Labeling}},
  booktitle = {\href{https://arxiv.org/abs/2410.14686}{DGON Positioning and Navigation for Intelligent Transport Systems (POSNAV)}},
  month = oct,
  year = {2024}
}

For further related publications, see:

@inproceedings{ott_heublein_icl,
  author = {Felix Ott and Lucas Heublein and Nisha Lakshmana Raichur and Tobias Feigl and Jonathan Hansen and Alexander Rügamer and Christopher Mutschler},
  title = {{Few-Shot Learning with Uncertainty-based Quadruplet Selection for Interference Classification in GNSS Data}},
  booktitle = {\href{https://ieeexplore.ieee.org/document/10578525}{IEEE Intl. Conf. on Localization and GNSS (ICL-GNSS)}},
  month = jun,
  year = {2024},
  address = {Antwerp, Belgium},
  doi = {10.1109/ICL-GNSS60721.2024.10578525}
}

@inproceedings{raichur_heublein,
  author = {Nisha L. Raichur and Lucas Heublein and Tobias Feigl and Alexander Rügamer and Christopher Mutschler and Felix Ott},
  title = {{Bayesian Learning-driven Prototypical Contrastive Loss for Class-Incremental Learning}},
  booktitle = {\href{https://arxiv.org/html/2405.11067v1}{arXiv preprint arXiv:2405.11067v1}},
  month = jul,
  year = {2024}
}

Acknowledgment

This work has been carried out within the DARCII project, funding code 50NA2401, supported by the German Federal Ministry for Economic Affairs and Climate Action (BMWK), managed by the German Space Agency at DLR and assisted by the Bundesnetzagentur (BNetzA) and the Federal Agency for Cartography and Geodesy (BKG). Additionally, this work was supported by the Bavarian Ministry of Economic Affairs, Regional Development and Energy through the Center for Analytics – Data – Applications (ADA-Center) within the framework of „BAYERN DIGITAL II“ (20-3410-2-9-8).

License

This work is licensed under a CC BY-NC-SA 4.0: Creative Commons Attribution-Noncommercial-ShareAlike, see https://creativecommons.org/licenses/by-nc-sa/4.0/

Kontakt

If you have any questions or tips to improve the datasets, contact us:

Felix Ott: felix.ott@iis.fraunhofer.de

Lucas Heublein: lucas.heublein@iis.fraunhofer.de

Nordostpark 84, 90411 Nürnberg, Germany, GoogleMaps

Fraunhofer Institute for Integrated Circuits IIS, Nürnberg, Germany