The dataset, assembled by a team from the Silesian University of Technology, Opole University of Technology, AGH University of Kraków, KP Labs, QZ Solutions and the European Space Agency’s ϕ-Lab, contains 1,954 hyperspectral image patches covering more than 43,000 hectares of southern Poland. That translates to roughly 108 million labeled pixels, each carrying 430 spectral bands spanning the visible, near-infrared and short-wave infrared ranges from 414 to 2,357 nanometers. According to the authors, it is the largest and most heterogeneous collection of its kind for bare soil detection, and the first to provide pixel-level annotations for entire agricultural parcels rather than isolated, spatially unmoored pixels.

The imagery was acquired on 3 March 2021 under cloudless, windless skies, using a HySpex VS-725 system flown aboard a Piper PA-31 Navajo aircraft at altitudes between 2,550 and 2,700 meters. Two sensors worked in tandem: a VNIR-1800 capturing 186 bands at 3.26 nanometer resolution across 400 to 1,000 nanometers, and a SWIR-384 capturing 288 bands at 5.45 nanometer resolution across 930 to 2,500 nanometers. The resulting ground sampling distance of 2 meters is fine enough to resolve individual field boundaries, hedgerows and dirt tracks, which matters enormously when the goal is to delineate whole fields rather than rough statistical averages.

What makes HyBEAR genuinely novel is its labeling philosophy. Most existing collections treat bare soil detection as a pixel-by-pixel classification problem, which can mislead downstream applications. Fertilization planning, tillage monitoring and erosion assessment all operate at the level of entire fields, so what farmers and agronomists actually need is an algorithm that says: this parcel, as a whole, is free of vegetation. HyBEAR’s annotators therefore labeled complete parcels, applying a strict rule that a field qualified as SOIL only when at least roughly 85 percent of its surface showed visible bare ground with no signs of mature vegetation. Ambiguous cases were flagged as MAYBE-SOIL and resolved through consensus among three experts with one, four and ten years of remote sensing experience, with the most senior scientist arbitrating the hardest calls.

Getting those labels right was harder than it sounds. Vegetation indices such as NDVI, which flag chlorophyll activity, can miss sparse or freshly emerging plants, while roads, rooftops and artificial surfaces often show equally low index values and can be mistaken for soil. Plain RGB composites proved similarly treacherous, since bare earth, dirt roads and shadowed ground can look nearly identical depending on flight altitude and sun angle. The team combined automated index analysis with careful visual inspection of both natural-color and color-infrared renderings, explicitly excluding dirt roads, infrastructure, deep shadows near tree lines and areas where crop residues complicated the spectral signature. The result is a ground truth designed to survive contact with messy reality.

Heterogeneity was built in deliberately. The two source scenes, P1 near the village of Przeworno in Lower Silesia and P2 south of Głubczyce in the Opolskie region, lie more than 60 kilometers apart yet were flown within an hour of each other. Because the sun and clouds shifted between acquisitions, lighting and reflectance conditions differ measurably between the two areas, forcing any trained model to generalize rather than memorize. The 250-by-250-pixel patches were divided into five spatially disjoint folds, and the benchmark prescribes a five-fold cross-validation protocol in which each fold serves once as the test set. This design directly probes whether an algorithm tuned on one landscape can transfer its knowledge to another, a question that haunts every operational Earth observation system.

To give future researchers a starting line, the authors ran six classic machine learning models through the protocol: logistic regression, L2-regularized linear models, AdaBoost, linear-kernel support vector machines, decision trees and random forests, each operating on the full 430-dimensional spectral vector of every pixel. The results were revealing. Logistic regression and support vector machines led the pack, with average accuracy around 0.926 to 0.927 and F-scores approaching 0.9. More striking was what happened on Fold 0, the only fold drawn from the P1 scene. Non-linear models such as random forests, AdaBoost and decision trees, which had apparently overfitted the spectral character of the larger P2 scene, degraded sharply when tested across locations, while the simpler linear models held up. In machine learning, the humblest method winning is often the most instructive outcome.

The failure cases are as informative as the successes. Qualitative inspection of patches like IMG_0022_F0 shows that errors cluster around physical confounders: tree shadows that mimic dark soil, dirt roads that share its texture, and crop residues that spectrally mix with the underlying ground. Fold 0 also carries the lowest soil pixel ratio in the dataset, at 28.3 percent in the full version, compounding the difficulty. The authors argue that these weaknesses point toward deep learning approaches capable of learning spatial context and richer representations, rather than judging each pixel in isolation. They also note that the benchmark’s parcel-level annotations remain largely untapped by the current pixel-wise baselines, leaving a clear opening for field-level segmentation methods.

Beyond agriculture, HyBEAR has an unexpected role to play in spacecraft engineering. As hyperspectral instruments proliferate on satellites, the bottleneck is no longer acquisition but transmission and on-board computing. The authors describe bare soil detection as a form of smart data compression: by pruning everything that is not soil before downlink, a satellite can slash the volume of data it ships to the ground and reserve its limited memory and compute for pixels that actually matter. Extracting soil parameters from non-soil pixels would produce noisy, inherently incorrect estimates anyway, so filtering early is both an efficiency gain and an accuracy safeguard. Related work from the team, including in-orbit vegetation index detection on the Intuition-1 mission and even quantum-kernel support vector machines, suggests this pipeline is already moving toward flight hardware.

The benchmark ships with reproducibility baked in. Alongside the imagery and labels, the Zenodo release includes wavelength metadata for all 430 bands, Jupyter notebooks covering preprocessing, normalization and the full cross-validation logic, and the ten trained baseline models so that anyone can independently verify the reported numbers. The authors are candid about limitations: the data represents a single early-spring snapshot of one region, so global soil diversity and seasonal variation remain untested, and the evaluation primarily addresses regional rather than worldwide generalization. Future expansions are planned to bring in more geographic zones and timeframes. Even so, HyBEAR fills a gap the community has long acknowledged, replacing ad hoc comparisons with a shared yardstick, and giving the fast-growing field of AI-driven soil mapping something it has rarely had: a fair, expert-verified and genuinely difficult test.

Subject of Research: A large-scale hyperspectral benchmark dataset with expert-verified parcel-level annotations for bare soil detection in precision agriculture

Article Title: HyBEAR: a hyperspectral benchmark for bare soil detection

Article References: Wijata, A. M., Ruszczak, B., Niepala, A., Gumiela, M., Smykala, K., Longépé, N., & Nalepa, J. (2026). HyBEAR: a hyperspectral benchmark for bare soil detection. Earth System Science Data, 18(10), 7301-7317. https://doi.org/10.5194/essd-18-7301-2026

Image Credits: AI Generated

DOI: 10.5194/essd-18-7301-2026

Keywords: hyperspectral imaging, bare soil detection, precision agriculture, remote sensing, machine learning benchmark, Earth observation, soil mapping, cross-validation, satellite data compression, agricultural datasets, HyBEAR, hyperspectral

The dataset, described in the journal Results in Engineering by Qing Li and Hao Xu, documents experiments on a rotor-bearing-gearbox system in which five distinct blade conditions were tested: a healthy blade and four seeded faults, namely a groove, wear, a partially broken section and a bent blade. Each condition was recorded under a stable speed of 20 Hz and under four non-stationary scenarios: continuously increasing speed, continuously decreasing speed, increasing then decreasing speed, and decreasing then increasing speed. The result is a matrix of 25 datasets covering every combination of health state and speed trajectory, all synchronously sampled at 20 kHz together with the instantaneous rotational speed signal.

The motivation is straightforward but consequential. Incipient blade faults such as small cracks, grooves or wear patches change the local mass distribution and stiffness of a rotor, disrupting its circumferential symmetry. Under steady conditions these faults produce characteristic, well-understood signatures. But when the shaft speed is sweeping continuously, the fault frequencies smear across the spectrum, a phenomenon known as spectral smearing, and the diagnostic features that work beautifully at constant speed can dissolve into noise. Machines in the field almost always operate in these non-stationary regimes, which means algorithms validated only on constant-speed data may fail precisely when they are needed most, during start-up, shutdown and load transients, when the risk of resonance-driven failure is elevated.

The experimental platform itself is a carefully instrumented drivetrain. A three-phase asynchronous motor, governed by a variable frequency drive, turns a shaft through a coupling into double-support bearings of type PH205, carrying a 120 mm aluminium alloy rotor disc fitted with six blades. Thirty-six M5 threaded holes, spaced at 10-degree intervals around the disc, allow precise dynamic balance adjustment. Downstream, a reducer gearbox connects to a magnetic powder brake, whose adjustable input current lets the researchers impose a controlled resistive load anywhere from 0 to 6 Nm. For all the runs in this dataset the brake was held at a fixed 4 Nm, isolating speed variation as the changing experimental factor.

The four fault types are quantitatively defined rather than vaguely described, which matters enormously for reproducibility. The groove fault is a cut 3 mm wide and 35 mm long; the wear fault removes material to a depth of 1.5 mm; the broken fault removes a 35 mm region of the blade; and the bending fault deforms the blade by 10 degrees. The healthy reference blade is 2 mm thick, 60 mm long and set at a 30-degree blade angle. Because each fault is a precisely specified geometric modification, other laboratories can in principle reproduce identical damage on their own rigs and compare results directly, something rarely possible with naturally occurring faults.

Each speed scenario was executed with care. In the stable-speed runs, 60 seconds of data were collected at 20 Hz and split into two 30-second parts. The ramping runs varied in duration, from roughly 15 seconds for a simple deceleration to nearly 39 seconds for a full up-down sweep, with speeds ranging between about 9 Hz and 20.75 Hz depending on the trajectory. Multiple independent repeated trials were conducted for each health condition and speed profile, with the fault condition, sensor arrangement, load and prescribed speed curve held identical across repetitions. This repetition strategy reduces the influence of random experimental disturbances and measurement variability, and gives users a basis for assessing the consistency of vibration responses under nominally identical conditions.

The sensing arrangement captures three-axis acceleration, with channels in the X, Y and Z directions recorded from an accelerometer mounted on the right side of the rotor disc, alongside the input velocity signal. Each data file is organised into five columns: operating time, acceleration in X, acceleration in Y, acceleration in Z, and rotational speed. A systematic file-naming convention encodes the health condition, the speed scenario and the dataset number, so that a file labelled 2_3_Groove_Up_Down, for instance, unambiguously identifies a grooved blade run through an increasing-then-decreasing speed profile. This kind of disciplined organisation is unglamorous, but it is exactly what turns a pile of recordings into a usable benchmark.

The authors also demonstrate one analytical workflow that the dataset supports: envelope order spectrum analysis, a technique built to defeat spectral smearing under variable speed. The method band-pass filters the vibration signal, applies a Hilbert transform to extract the amplitude envelope of the modulated carrier, resamples the envelope uniformly in the angular domain using the integrated instantaneous rotational frequency, and then applies a discrete Fourier transform to produce a spectrum expressed in orders of the shaft rotation rather than in hertz. Because the angular domain is tied to shaft position rather than wall-clock time, features that would blur in a conventional frequency spectrum remain sharp regardless of how fast the shaft is turning.

Applied to the increasing-speed run of the grooved blade, the technique is revealing. The band-pass filtered signal shows periodic fluctuations confirming continuous pulse excitation from the damaged blade, and the overall envelope amplitude grows with time as the rising speed injects more impact energy. In the resulting envelope order spectrum, distinct peaks appear at 1.00X, 2.01X, 4.02X and 7.03X, matching the theoretical vibration features of a grooved blade, while noise components stay at a low level. The physics is intuitive: the local material removal disrupts the circumferential symmetry of the rotor, producing a periodic excitation once per revolution that yields the dominant 1X component, while the non-sinusoidal character of the excitation, filtered through the structural transfer path and nonlinear response of the system, generates the higher-order harmonics.

The research landscape that motivated this release is telling. The celebrated Case Western Reserve University bearing dataset, the Paderborn University bearing collections, the IEEE PHM 2012 run-to-failure bearing data, the PHM 2009 gearbox challenge data and the Tsinghua University gearbox datasets have collectively underpinned two decades of progress in data-driven diagnostics, yet nearly all were recorded at constant speed or at discrete speed steps, and almost none address rotor-blade damage specifically. By publishing labelled blade-fault data under genuinely continuous speed variation, the AAU team opens the door to benchmarking speed-conditioned feature extraction, cross-condition fault diagnosis, self-supervised learning and uncertainty-aware diagnostic methods on problems that actually resemble field conditions. The authors are candid about limitations: only five health states are covered, laboratory safety precludes the 3000 to 5000 rpm range of real turbomachinery, and the brake load is fixed. Future extensions, they note, will broaden the damage modes, widen the speed range, add adjustable loading, nonlinear ramps and random speed fluctuations. Even so, for a field long starved of non-stationary rotor data, this release is a welcome and overdue gift.

Subject of Research: A public rotor-blade vibration dataset for fault diagnosis under time-varying rotational speed conditions

Article Title: AAU Rotor Vibration Dataset Under Time-Varying Speed Scenarios

Article References: Li, Q., & Xu, H. (2026). AAU Rotor Vibration Dataset Under Time-Varying Speed Scenarios. Results in Engineering, 32, Article 113172. https://doi.org/10.1016/j.rineng.2026.113172

Image Credits: AI Generated

DOI: 10.1016/j.rineng.2026.113172

Keywords: rotor blades, vibration monitoring, fault diagnosis, time-varying speed, open dataset, envelope order spectrum, predictive maintenance, rotating machinery, machine learning benchmark, signal processing, Rotor, Vibration