Organizations: College of Intelligence Science and Technology, National University of Defense Technology, Changsha, Hunan 410073, China · College of Electrical and Information Engineering, Hunan University, Changsha 410082, China
Passive long-wave infrared (LWIR) hyperspectral ranging enables distance estimation in low-light and nighttime scenes by exploiting atmospheric absorption features in thermal radiance received through the atmosphere.Joint estimation of temperature, emissivity, and distance is computationally expensive. Reference-range joint inversion also uses a distance-invariant effective attenuation coefficient, which can bias range estimates.We introduce transmittance extraction and distance alignment (TEDA), which decouples range estimation from temperature--emissivity inversion. In the first stage, a baseline estimator with a data-fidelity term invariant to the known absorption direction yields two closed-form smoothing branches for the slowly varying thermal continuum. An observation-derived gate combines the branches, and subtracting the blended baseline in the log domain recovers atmospheric transmittance. The second stage estimates range by matching the recovered transmittance to sensor-domain transmittance models recomputed for each candidate distance. Monte Carlo simulations show that TEDA effectively reduces the ranging bias caused by the distance-invariant attenuation coefficient approximation. In a measured scene, TEDA's mean range estimates are closer to the LiDAR medians than those of reference-range joint inversion in both evaluated patches. TEDA processes a complete 256×256 region of interest in 8.19s versus 159.47s for reference-range joint inversion, an approximately 20-fold speedup.
Figures & tables
Fig. 1: Transmittance extraction and distance alignment (TEDA) for passive LWIR hyperspectral ranging. A push-broom camera scans the scene through an absorbing air path, so each measured spectrum contains range-dependent atmospheric transmittance. The red-boxed tree pixel illustrates the two-stage processing applied to every pixel. In Stage I, the known absorption structure guides the separation of atmospheric transmittance from the slowly varying thermal background. In Stage II, the extracted transmittance is aligned over the water-vapor band against model spectra that account for the sensor response throughout the distance search. The output shows the hyperspectral distance estimate alongside the LiDAR reference map; the histogram compares their distance distributions over an 8×8 patch around the marked pixel. Molecular absorption is computed from HITRAN2020 for a model atmosphere with water-vapor volume mixing ratio (VMR) 0.012, temperature 289.7 K, and pressure 1010 mbar. The instrumental spectral response function (ISRF) is Gaussian with a full width at half maximum (FWHM) of 40 nm.
Fig. 2: Comparison of continuous-spectrum water-vapor transmittance with the transmittance after spectral-response integration. (a) Full 8–13 μ m spectrum; the shaded intervals mark the ranges enlarged below. (b) The 8–9 μ m absorption structure. (c) The 12–13 μ m absorption structure. The panels compare the continuous-spectrum transmittance and the spectrally integrated result. The air temperature is 289.7 K, the pressure is 1010 mbar, the water-vapor volume mixing ratio is 0.012, and the propagation distance is 200 m. The sensor ISRF has a 40 nm full width at half maximum.
Fig. 3: Quantification of the distance-invariant attenuation coefficient approximation. (a) Transmittance at 50, 100, and 150 m. The rows correspond to the three distances; the columns show the full sensor band and magnified views of 8.0–8.9 μ m and 12.3–13.0 μ m. The solid and dashed curves denote the ISRF-integrated and approximate transmittances, respectively. The approximation derives an effective attenuation coefficient at a reference distance of 200 m and applies Beer–Lambert scaling without recalculating the spectral-response integral. (b) Distance-estimation bias of the approximate transmittance model as a function of the true distance. The bias vanishes at the reference distance of 200 m and at zero distance and attains its maximum at an intermediate distance, consistent with the convexity bound ( 15 ).
Fig. 4: Effect of excluding the absorption direction from the Stage-I data-fidelity term. (a) The Euclidean data-fidelity term allows a smooth baseline to move toward the observation along the absorption direction, causing leakage. (b) The projected data-fidelity term is invariant along this direction, so the corresponding baseline component is set by smoothness alone. (c) Absolute Pearson correlation defined in ( 33 ). Boxes summarize the correlation between the true thermal-continuum baseline and the absorption prior across emissivity spectra from the ECOSTRESS spectral library for a target 5 K colder than the air. The dashed line gives the same statistic for the sensor-domain log-transmittance, the atmospheric absorption term. (d) Noise-free range bias at 100 m under operator-consistent alignment. Both PLS estimators use the same operator-consistent distance-matching procedure; the estimator with M=I−αα⊤ is the absorption-prior method used in Stage I of TEDA. Boxes show the interquartile range and whiskers the 5th–95th percentiles.
Figure 5
Method
−8 K
−5 K
−2 K
Time (s)
NasPLS [ 39 ]
157.8
190.3
192.1
2.157
arPLS [ 18 ]
28.1
31.3
44.0
1.637
Reference-range joint inversion [ 16 ]
20.1
20.7
20.0
19.57
Per-range joint inversion
4.8
6.6
10.3
943.02
TEDA (proposed)
3.9
5.3
8.7
1.571
Oracle baseline
2.0
2.8
4.5
–
TABLE I: Distance RMSE (m) and batch runtime (s) comparison of six ranging methods on noisy simulated measurements of a rock at 100 m; 100 noise realizations per temperature difference, times include model-bank construction; fits and transmittance estimates shown in Figs. V-A and V-A .
Configuration
−8 K
−5 K
−2 K
Full method
3.9
5.3
8.7
w/o absorption-direction projection
4.3
5.9
9.2
w/o operator-consistent bank
9.4
9.6
11.1
Uniform matching weights
4.0
5.5
9.1
Single smoothing strength, η=2
4.4
6.1
10.0
Single smoothing strength, η=16
3.7
5.2
8.5
TABLE II: Effect of removing or replacing each TEDA component on ranging error, on the same rock-target simulation at 100 m as Table I ; distance RMSE (m) over 100 noise realizations per temperature difference.
Configuration
Rock
Mineral
Soil
Vegetation
Overall
Single η=2
9.8
9.6
10.3
9.2
9.7
Single η=16
14.0
12.7
14.3
8.4
12.6
Dual η=2/16
9.7
9.3
10.3
8.5
9.5
TABLE III: Distance RMSE (m) of three TEDA smoothing-strength configurations on four material categories; 1,528 library spectra at 50, 100, and 200 m and −8 , −5 , and −2 K, 100 common-noise realizations per condition, and the overall column category-balanced.
Fig. 7: Full-frame passive ranging result and LiDAR reference. (a) Distance map estimated by TEDA from the hyperspectral measurements. (b) LiDAR range map shown in the hyperspectral image coordinates for visual comparison. Both panels use the same fixed 20–200 m color scale; valid distances outside this interval are clipped to the endpoint colors. The two maps exhibit consistent near-to-far ordering across the grassland, foreground tree and background forest.
Fig. 8: Comparison of passive ranging methods in a 256×256 region. (a)–(d) Distance maps estimated by arPLS, reference-range joint inversion, per-range joint inversion, and TEDA, respectively. The red dashed line marks column 20, and markers (f)–(h) identify the grass, foreground-tree, and background-tree evaluation patches. (e) Vertical distance profiles along the marked column, indexed from the top to the bottom of the region. (f)–(h) Distance distributions within the three marked 8×8 patches; the light shading in (e) marks contiguous pixel spans with valid LiDAR observations. The profile and distributions show that TEDA retains the range layering recovered by per-range joint inversion while avoiding the systematic underestimation of arPLS and reference-range joint inversion.
Method
Grass
Foreground tree
Background tree
Time (s)
arPLS
21.1±2.2
38.2±8.9
100.1±18.9
58.84
Reference-range joint inversion
28.8±2.0
47.5±4.2
84.5±6.8
159.47
Per-range joint inversion
31.6±4.8
66.5±12.5
151.8±17.3
24076.50
TEDA (proposed)
30.0±2.8
60.2±7.9
153.8±17.1
8.19
LiDAR reference (median)
31.7
62.1
–
–
TABLE IV: Mean and standard deviation of the distance estimates of four ranging methods in three 8×8 patches of the measured scene (grass, foreground tree, background tree; marked in Fig. 8 ); the LiDAR median is the reference, and the runtime covers the full region.
Fig. 9: Passive ranging results on additional DARPA Invisible Headlights scenes. Each column shows one acquisition group. (a) LWIR spectral-mean images. (b) LiDAR range references. (c) TEDA distance estimates. All scenes are processed with the same fixed hyperparameters, without per-scene tuning.
Long-wave infrared (LWIR) hyperspectral observations contain distance-dependent atmospheric absorption signatures, providing a physical basis for long-range passive ranging. However, in natural scenes, these signatures are nonlinearly coupled with target temperature, material emissivity, and path radiance, making distance inversion from observed radiance ill posed. Existing methods typically rely on full-band measurements and pixel-wise joint optimization, which is computationally expensive and does not explicitly exploit sharp atmospheric absorption structures. This paper proposes an Absorption-Guided Distance-Decoupled Estimation and Refinement (ADER) framework for LWIR hyperspectral passive ranging. ADER represents emissivity with B-spline control points under a smoothness prior, suppressing overfitting to atmospheric absorption structures and enabling distance-decoupled estimation. It further uses ozone-absorption cues to classify pixels into emission-dominant and reflection-dominant groups. For emission-dominant pixels, ADER compensates path radiance and transmittance and estimates distance by one-dimensional absorption-residual minimization. For reflection-dominant pixels, ADER refines the initial estimate using downwelling-radiance compensation based on the complete radiative model. To reduce spectral redundancy, ADER also introduces a greedy band selection strategy based on multi-scene effective Fisher information for the distance parameter. Experiments on real scenes show that ADER recovers LiDAR-consistent spatial distance structures under both full-band and 20-band settings, improves ranging accuracy in the evaluated regions, and achieves approximately two orders of magnitude speedup over a public full-band hyperspectral ranging method.
Shuo Liu, Chen Fan, Zhihe Chen +2
College of Intelligence Science and Technology, National University of Defense Technology, Changsha, China
Passive long-wave infrared (LWIR) hyperspectral imaging under a standoff geometry depends on atmospheric absorption and emission, as well as reflected radiance, thus making atmospheric compensation essential to get knowledge of a target of interest. Despite its importance, this compensation has been largely overlooked due to its practical and modeling difficulty. In this paper, we present a lightweight set-based deep learning framework that takes multiple radiance measurements, collected at different standoff ranges, as input and jointly estimates transmittance, atmospheric path radiance, and a shared downwelling spectrum. We analyze the learned representation with a sparse autoencoder and observe that several latent features do activate on geographically coherent subsets of the test data despite the absence of location supervision. Experiments on a MODTRAN generated standoff LWIR dataset demonstrate low spectral distortion across all estimated products. The dataset and code is publicly available at: https://factral.co/SAE-LWIR/
Fabian Perez, Nicolas Quintero, Jeferson Acevedo +1
Department of Computer Science, Universidad Industrial de Santander
Temperature-emissivity-texture (TeX) decomposition seeks to recover object heat state, material spectral response, and visible-like geometric texture from long-wave infrared hyperspectral imaging (LWIR HSI). Existing TeX pipelines are mainly scene-specific inverse solvers, and the lack of paired LWIR HSI-TeX supervision has limited learning-based decomposition. To address this gap, we introduce TeX-1500, a large-scale paired LWIR HSI-TeX dataset and benchmark for supervised HSI-to-TeX decomposition. TeX-1500 contains 1,522 calibrated real-scene pairs from DARPA Invisible Headlights (DARPA IH) pushbroom imagery and our FTIR acquisitions, covering five locations, four seasons, diverse acquisition times, heterogeneous wavelength layouts, and two sensor families. Each sample stores a calibrated valid-band radiance cube, calibrated wavelength positions, and aligned temperature, emissivity, and texture supervision constructed through a consistent restoration and TeX-construction protocol. We further provide TeX-UNet, a simple wavelength-aware baseline that maps calibrated HSI bands and wavelength positions to TeX fields. Experiments on the held-out DARPA IH pushbroom scenes and zero-/few-shot transfer to FTIR scenes show that TeX-1500 provides usable paired supervision and a measurable benchmark for data-driven physical-property-centered thermal perception.
Cheng Dai, Jiale Lin, Hongyi Xu +3
School of Science, Westlake University, Hangzhou 310030, China · School of Engineering, Westlake University, Hangzhou 310030, China