From Data to Model: Anomaly Detection of 3D GPR Data in CuDeRes Model Space

Xiren Zhou, Shikang Liu, Ao Chen, Xinyu Yan, Huanhuan Chen

Keywords: Cubic Decay Reservoir Network (CuDeRes), Ground Penetrating Radar (GPR), Learning in the Model Space, Reservoir Computing

TL;DR: We propose anomaly detection of GPR data in the Cubic Decay Reservoir Network (CuDeRes) model space, effectively and accurately capturing multi-directional dynamics and enabling accurate anomaly detection and clustering even with limited normal data.

Abstract: Ground Penetrating Radar (GPR) offers in-depth subterranean insights, yet subsurface anomaly detection in GPR data remains challenging due to limited training data, typically confined to some normal data samples free from any subsurface structures or anomalies, and the variability of subsurface conditions. In response, this paper introduces practical and accurate subsurface anomaly detection within the Cubic Decay Reservoir Network (CuDeRes) model space. Our approach employs commonly available normal GPR data, segmented into blocks. Each data block is independently fitted using the introduced CuDeRes, which incorporates three reservoirs with spatial decay to adequately capture the data-inherent multi-directional dynamics, resulting in a compact fitted readout model. Representing each data block with the fitted model, together with the distance measurement between models, the original GPR data blocks are mapped into the CuDeRes model space, and the fitted models are collected into a "Model Depot". For subsequent anomaly detection in newly collected GPR data, the same segmentation and CuDeRes fitting approaches are applied, where the data blocks are represented by fitted models for comparative assessment against the model depot. Anomalies are detected through model dissimilarities, and subsequently clustered within the CuDeRes model space, allowing us to accurately identify the data blocks with potential subsurface anomalies and ascertain their anomaly types. Experiments on real-world GPR data demonstrate the practical effectiveness of our approach, notably using only limited normal data.

Primary Area: applications to physical sciences (physics, chemistry, biology, etc.)

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.

Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 2100