GPR Datasets for Machine Learning and AI
Real-world Ground Penetrating Radar data for deep learning, computer vision and automated interpretation research
Download real-world Ground Penetrating Radar datasets for machine learning, deep learning, computer vision, signal processing and automated GPR interpretation. Every dataset on GPRbase contains raw radargrams recorded on real sites — with the authentic noise, clutter, coupling effects, and target variability that make GPR interpretation a genuinely hard machine learning problem.
Automated GPR interpretation is an active research field: hyperbola detection, rebar localization, utility classification, layer picking, void detection, and end-to-end radargram segmentation. Progress in all these tasks depends on access to real, diverse training data. Synthetic simulators like gprMax are valuable for pre-training and data augmentation, but models trained purely on synthetic data notoriously fail to generalize to field conditions. GPRbase exists to close that gap.
All datasets are free, raw (no processing applied), documented with antenna and frequency metadata, and distributed under Creative Commons BY-NC-SA 4.0 — usable for academic research, including publications, provided derivatives are shared alike. Download the data below, and if your lab produces GPR acquisitions, consider contributing to grow the shared pool.
Datasets for ML research
Raw radargrams with documented acquisition parameters — download and explore — 17 datasets available
Confirmed Karst Cavities in Limestone — 300 MHz GPR Profiles to 4 m Depth
Raw GPR profiles acquired over limestone containing confirmed karst cavities, using a GSSI UtilityScan DF dual-frequency antenna. The 300 MH…
Reinforced Concrete Slab with Expansion Joint
GPR dataset containing 7 files collected on a reinforced concrete slab using a GSSI FLEX NX structure-scanning radar. The data clearly show …
Utilities & Interfaces in Favorable Soil with Deep GPR Penetration
GPR dataset containing 1 file collected with a GSSI UtilityScan DF dual-frequency 300/800 MHz ground penetrating radar in highly favorable s…
Ground-penetrating radar detection of gas connections (HDPE polyethylene)
4 dual-frequency ground-penetrating radar profiles (UtilityScan DF) showing the location of HDPE (High-Density Polyethylene) gas connections…
3D concrete scanning over a slab-access ramp interface
60 x 60 cm 3D grid (10 cm spacing). Starting point at bottom left (0,0) and profiles from left to right, then from bottom to top. Try to rec…
Detection of various buried utilities using a dual-frequency antenna (300 MHz and 800 MHz) Highway pavement inspection along an emergency stopping lane (ESL)
One GPR profile across an avenue, crossing several buried utilities. How many do you see?…
Two layers of rebars in a reinforced concrete structure (3D grid)
26 profiles (60 x 60cm grid), 5cm spacing. Ideal dataset for practicing 3D processing with GPRviewer, and easier to see the best possible re…
Roadway inspection along an emergency stopping lane (ESL)
Six dual-frequency ground-penetrating radar (GPR) profiles (300/800 MHz) showing numerous interfaces and pavement structure layers in a soil…
Thickness detection above three masonry bridge arches
Ground-penetrating radar measurements recorded with UtilityScan DF 300/800 MHz from the roadway…
Detection of buried pipes with the NX15 step-frequency ground-penetrating radar antenna
High resolution imaging with the NX15 antenna, for detecting buried cables and pipes. Observe the data quality, ideal for locating small-dia…
Voids in a concrete block wall
FLEX NX file, dual antenna (inline & cross-line), created on a concrete block wall with gaps (no concrete) at the end of the profile. Test t…
3D Grid Concrete bloc wall with voids
3D grid 60x60 cm created with the FLEX NX radar, 3D mode (26 profiles per antenna) Spacing : 5 cm Origin (0,0) : bottom-left…
Georeferenced dual-frequency ground-penetrating radar (GPR RTK) files
16 Radar profiles in a parking lot, showing heterogeneous soil conditions. The data is georeferenced and can be imported into GPRviewer for …
FLEX, NX25 and NX15 reference lines (MDSLab)
Three radar profiles were created on the same line, one of our reference lines in the MDSLab in Aix-en-Provence. Compare the measurements of…
Double layer of reinforcement and structural change (1 profile)
One radargram. Two identifiable steel layers, the second being more difficult to discern. A change in structure is also visible before 50 cm…
FLEX 3D grid 60 x 60 cm (reinforced concrete beam and slab).
INLINE (channel 1) and CROSS-LINE (channel 2) profiles: 26 profiles, 13 in X, 13 in Y (5 cm spacing)…
Detection of a backfilled marl pit beneath a road
This profile shows a measurement taken at the level of a roadway, with the presence of a marl pit that has been filled in. Several layers of…
Why real data matters for GPR AI
Models trained exclusively on synthetic radargrams learn clean hyperbolas over homogeneous backgrounds — then fail on field data full of surface clutter, antenna ringing, soil heterogeneity, and overlapping reflections. Real datasets expose your model to this complexity from the start. The most robust published approaches combine synthetic pre-training with fine-tuning on real data: GPRbase provides the real component of that recipe.
Machine learning tasks on GPR data
Common research problems include: hyperbola detection and fitting (object localization), rebar detection and cover estimation in concrete, buried utility detection and material classification, layer boundary picking for pavements and stratigraphy, void and defect detection, and full radargram semantic segmentation. Each GPRbase dataset documents its targets, letting you select data matching your task — rebar-rich concrete scans for detection models, multi-utility corridors for classification, layered pavements for boundary picking.
Data structure: from DZT to tensors
A radargram is naturally a 2D array: traces (columns) × time samples (rows), with 16 or 32-bit amplitude values. Python libraries like readgssi load DZT files directly into numpy arrays, ready for PyTorch or TensorFlow pipelines. B-scans can be treated as images for CNN-based approaches, as sequences for recurrent models, or as raw signals for 1D convolutional architectures. The DZT header provides the physical calibration (time range, samples per scan) needed to convert predictions back to depths.
Preprocessing and augmentation
Typical preprocessing includes time-zero correction, background removal (mean trace subtraction), gain compensation, and normalization. Because GPRbase data is completely raw, you control the full pipeline — essential for reproducible research and for studying preprocessing sensitivity. For augmentation, real data can be combined with synthetic radargrams, cropped into patches, flipped horizontally, and noise-injected. Multi-antenna variety across datasets naturally augments frequency diversity.
Generalization across sites and antennas
A model that works on one site with one antenna often fails elsewhere. GPRbase datasets span multiple applications (concrete, utilities, geotechnical, archaeology, roads, geosciences), antennas (from 100 MHz to 2.7 GHz), and site conditions. This variety enables meaningful cross-domain evaluation: train on some datasets, test on held-out sites, and measure true generalization rather than site memorization.
Citing and publishing with GPRbase data
CC BY-NC-SA 4.0 permits research use including peer-reviewed publications, with attribution. Cite the dataset identifier (dsXXX) and the GPRbase URL in your data availability statement. If you build annotated versions (bounding boxes, masks, picks), the share-alike clause means publishing them under the same license — growing the commons for the whole GPR ML community.
Frequently asked questions — GPR machine learning
Are these datasets annotated?
The datasets are raw radargrams with documented targets in the metadata, but without pixel-level annotations. Target descriptions (rebar, utilities, layers) guide your own labeling. Annotated derivatives shared under CC BY-NC-SA 4.0 are welcome contributions.
How do I load DZT files in Python?
The readgssi library reads DZT files into numpy arrays: pip install readgssi, then readgssi.readgssi(infile=path). GPRPy is an alternative with processing functions. From numpy, the data goes directly into PyTorch or TensorFlow.
Can I use these datasets in a published paper?
Yes, CC BY-NC-SA 4.0 permits academic publication with attribution. Cite the dataset identifier and GPRbase URL. Non-commercial use only; commercial licensing requires contacting info@gprbase.com.
Is the data enough to train a deep network from scratch?
For most deep architectures, combine it with synthetic pre-training (e.g., gprMax simulations) or self-supervised learning, then fine-tune on the real data. Patch-based training multiplies the effective sample count significantly.
Do the datasets share a common format?
Yes — all datasets are GSSI DZT files with consistent structure, which simplifies building a single data loader across the whole collection. Antenna and frequency metadata allow conditioning or filtering.
Working on GPR + AI? Contribute your data
Every contributed dataset grows the shared training pool for the entire research community. Raw acquisitions, annotated sets, and benchmark splits are all welcome.
Submit your data