Improving Deep Learning for Maritime Remote Sensing through Data Augmentation and Latent Space
Training deep learning models requires having the right data for the problem and understanding both your data and the models’ performance on that data. Training deep learning models is difficult when data are limited, so in this paper, we seek to answer the following question: how can we train a dee...
| Main Authors: | , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
MDPI AG
2022-07-01
|
| Series: | Machine Learning and Knowledge Extraction |
| Subjects: | |
| Online Access: | https://www.mdpi.com/2504-4990/4/3/31 |
| _version_ | 1827659394093940736 |
|---|---|
| author | Daniel Sobien Erik Higgins Justin Krometis Justin Kauffman Laura Freeman |
| author_facet | Daniel Sobien Erik Higgins Justin Krometis Justin Kauffman Laura Freeman |
| author_sort | Daniel Sobien |
| collection | DOAJ |
| description | Training deep learning models requires having the right data for the problem and understanding both your data and the models’ performance on that data. Training deep learning models is difficult when data are limited, so in this paper, we seek to answer the following question: how can we train a deep learning model to increase its performance on a targeted area with limited data? We do this by applying rotation data augmentations to a simulated synthetic aperture radar (SAR) image dataset. We use the Uniform Manifold Approximation and Projection (UMAP) dimensionality reduction technique to understand the effects of augmentations on the data in latent space. Using this latent space representation, we can understand the data and choose specific training samples aimed at boosting model performance in targeted under-performing regions without the need to increase training set sizes. Results show that using latent space to choose training data significantly improves model performance in some cases; however, there are other cases where no improvements are made. We show that linking patterns in latent space is a possible predictor of model performance, but results require some experimentation and domain knowledge to determine the best options. |
| first_indexed | 2024-03-09T23:20:51Z |
| format | Article |
| id | doaj.art-bec0ec685d90495f8edad4634061fb3e |
| institution | Directory Open Access Journal |
| issn | 2504-4990 |
| language | English |
| last_indexed | 2024-03-09T23:20:51Z |
| publishDate | 2022-07-01 |
| publisher | MDPI AG |
| record_format | Article |
| series | Machine Learning and Knowledge Extraction |
| spelling | doaj.art-bec0ec685d90495f8edad4634061fb3e2023-11-23T17:28:09ZengMDPI AGMachine Learning and Knowledge Extraction2504-49902022-07-014366568710.3390/make4030031Improving Deep Learning for Maritime Remote Sensing through Data Augmentation and Latent SpaceDaniel Sobien0Erik Higgins1Justin Krometis2Justin Kauffman3Laura Freeman4National Security Institute, Virginia Tech, Arlington, VA 22203, USADepartment of Aerospace and Ocean Engineering, Virginia Tech, Blacksburg, VA 24061, USANational Security Institute, Virginia Tech, Arlington, VA 22203, USANational Security Institute, Virginia Tech, Arlington, VA 22203, USANational Security Institute, Virginia Tech, Arlington, VA 22203, USATraining deep learning models requires having the right data for the problem and understanding both your data and the models’ performance on that data. Training deep learning models is difficult when data are limited, so in this paper, we seek to answer the following question: how can we train a deep learning model to increase its performance on a targeted area with limited data? We do this by applying rotation data augmentations to a simulated synthetic aperture radar (SAR) image dataset. We use the Uniform Manifold Approximation and Projection (UMAP) dimensionality reduction technique to understand the effects of augmentations on the data in latent space. Using this latent space representation, we can understand the data and choose specific training samples aimed at boosting model performance in targeted under-performing regions without the need to increase training set sizes. Results show that using latent space to choose training data significantly improves model performance in some cases; however, there are other cases where no improvements are made. We show that linking patterns in latent space is a possible predictor of model performance, but results require some experimentation and domain knowledge to determine the best options.https://www.mdpi.com/2504-4990/4/3/31data augmentationdimensionality reductionlatent spaceUMAPsimulated datadeep neural network |
| spellingShingle | Daniel Sobien Erik Higgins Justin Krometis Justin Kauffman Laura Freeman Improving Deep Learning for Maritime Remote Sensing through Data Augmentation and Latent Space Machine Learning and Knowledge Extraction data augmentation dimensionality reduction latent space UMAP simulated data deep neural network |
| title | Improving Deep Learning for Maritime Remote Sensing through Data Augmentation and Latent Space |
| title_full | Improving Deep Learning for Maritime Remote Sensing through Data Augmentation and Latent Space |
| title_fullStr | Improving Deep Learning for Maritime Remote Sensing through Data Augmentation and Latent Space |
| title_full_unstemmed | Improving Deep Learning for Maritime Remote Sensing through Data Augmentation and Latent Space |
| title_short | Improving Deep Learning for Maritime Remote Sensing through Data Augmentation and Latent Space |
| title_sort | improving deep learning for maritime remote sensing through data augmentation and latent space |
| topic | data augmentation dimensionality reduction latent space UMAP simulated data deep neural network |
| url | https://www.mdpi.com/2504-4990/4/3/31 |
| work_keys_str_mv | AT danielsobien improvingdeeplearningformaritimeremotesensingthroughdataaugmentationandlatentspace AT erikhiggins improvingdeeplearningformaritimeremotesensingthroughdataaugmentationandlatentspace AT justinkrometis improvingdeeplearningformaritimeremotesensingthroughdataaugmentationandlatentspace AT justinkauffman improvingdeeplearningformaritimeremotesensingthroughdataaugmentationandlatentspace AT laurafreeman improvingdeeplearningformaritimeremotesensingthroughdataaugmentationandlatentspace |