A UAV-Based VNIR Hyperspectral Benchmark Dataset for Landmine and UXO Detection

This paper introduces a novel benchmark dataset of Visible and Near-Infrared (VNIR) hyperspectral imagery acquired via an unmanned aerial vehicle (UAV) platform for landmine and unexploded ordnance (UXO) detection research. The dataset was collected over a controlled test field seeded with 143 realistic surrogate landmine and UXO targets, including surface, partially buried, and fully buried configurations. Data acquisition was performed using a Headwall Nano-Hyperspec sensor mounted on a multi-sensor drone platform, flown at an altitude of approximately 20.6 m, capturing 270 contiguous spectral bands spanning 398-1002 nm. Radiometric calibration, orthorectification, and mosaicking were performed followed by reflectance retrieval using a two-point Empirical Line Method (ELM), with reference spectra acquired using an SVC spectroradiometer. Cross-validation against six reference objects yielded RMSE values below 1.0 and SAM values between 1 and 6 degrees in the 400-900 nm range, demonstrating high spectral fidelity. The dataset is released alongside raw radiance cubes, GCP/AeroPoint data, and reference spectra to support reproducible research. This contribution fills a critical gap in open-access UAV-based hyperspectral data for landmine detection and offers a multi-sensor benchmark when combined with previously published drone-based electromagnetic induction (EMI) data from the same test field.

💡 Research Summary

This paper presents a comprehensive, openly available benchmark dataset of visible‑near‑infrared (VNIR) hyperspectral imagery captured from an unmanned aerial vehicle (UAV) for the purpose of landmine and unexploded ordnance (UXO) detection research. The authors deployed 143 realistic surrogate landmine and UXO objects—covering surface, partially buried, and fully buried configurations—on a controlled test field in June 2023. One year later, in June 2024, a customized DJI Matrice 600 Pro equipped with a Headwall Nano‑Hyperspec® sensor was flown at an altitude of approximately 20.6 m to acquire VNIR data over the same area. The sensor operates as a 1‑by‑640 line scanner, recording 270 contiguous spectral bands from 398 nm to 1002 nm with an average spectral resolution of about 2.2 nm. Thirty‑two flight lines were executed, producing 32 raw digital‑number (DN) cubes that were converted to radiance using manufacturer‑provided calibration coefficients, orthorectified with Applanix GPS/IMU data, and mosaicked in ENVI Classic 5.7 into a single radiance orthomosaic.

Radiometric calibration to surface reflectance was performed using a two‑point Empirical Line Method (ELM). Two calibration panels (light gray and black) with known reflectance spectra measured by a Spectra Vista Corporation (SVC) spectroradiometer served as the reference points. For each spectral band, a linear relationship R = a·L + b was derived, where L is at‑sensor radiance and R is surface reflectance. The resulting reflectance cube was then georeferenced using a dense network of ground control points (GCPs) and AeroPoints in QGIS, yielding a final product of size 3123 × 6631 × 272 (lines × samples × bands). All raw radiance cubes, calibration panel spectra, GCP/AeroPoint coordinates, and the processed reflectance cube are released together.

To assess the fidelity of the ELM correction, six reference objects were selected: four calibration panels (light gray, medium gray, dark gray, black) and two in‑scene targets (PFM‑1 and an M65Al projectile). Reference spectra for these objects were obtained with the SVC spectroradiometer and resampled to match the image bands. The authors computed Root Mean Square Error (RMSE) and Spectral Angle Mapper (SAM) between the image‑derived spectra and the spectroradiometer measurements. Over the full 400‑1000 nm range, RMSE values ranged from 0.5 % to 4.5 % and SAM from 1° to 12°. When the analysis was limited to the higher‑signal 400‑900 nm window, RMSE dropped below 1 % and SAM fell between 1° and 6°, indicating excellent radiometric accuracy and spectral shape preservation in the most informative portion of the VNIR spectrum. Degradation beyond 900 nm was attributed to reduced sensor sensitivity, lower solar irradiance, atmospheric water‑vapor absorption near 940 nm, and the linearity assumption of the ELM under low‑signal conditions.

The paper positions this dataset as a critical resource for the remote‑sensing community. Compared with existing public datasets, it is the first to provide UAV‑borne VNIR hyperspectral data specifically for landmine detection, and it complements a previously published drone‑based electromagnetic induction (EMI) dataset collected over the same field. The combined multimodal suite enables researchers to explore sensor‑fusion strategies, develop and benchmark machine‑learning or deep‑learning algorithms for spectral target detection, anomaly detection, and classification, and evaluate the robustness of such methods under realistic environmental variations (vegetation growth, soil moisture, seasonal changes). The authors also supply detailed processing documentation, facilitating reproducibility and lowering the barrier for new entrants to the field.

Limitations are acknowledged: the VNIR sensor’s performance deteriorates above 900 nm, making the SWIR region unavailable; the two‑point ELM assumes linear atmospheric and sensor behavior, which may not hold under rapidly changing illumination or atmospheric conditions; and the dataset reflects a specific geographic and climatic context (mid‑western United States), so external validation on different soils, vegetation types, or climates will be necessary for broader generalization.

Future work suggested includes extending the acquisition to additional spectral ranges (SWIR, LWIR), collecting data under varied weather and seasonal conditions, and investigating more sophisticated atmospheric correction techniques (e.g., multi‑point ELM, physics‑based models). The authors also encourage the community to leverage the dataset for advancing safe, efficient, and cost‑effective demining technologies, ultimately contributing to humanitarian efforts worldwide.