Ozone Cues Mitigate Reflected Downwelling Radiance in LWIR Absorption-Based Ranging
Authors: Unay Dorken Gallastegi, Wentao Shangguan, Vaibhav Choudhary, Akshay Agarwal, Hoover Rueda-Chacón, Martin J. Stevens, Vivek K Goyal
Organizations: Department of Electrical and Computer Engineering, Boston University, Boston, MA 02215 USA · Boston University · Department of Computer Science, Universidad Industrial de Santander, Bucaramanga, 680002, Colombia · National Institute of Standards and Technology, Boulder, CO, 80305 USA
Passive long-wave infrared (LWIR) absorption-based ranging relies on atmospheric absorption to estimate distances to objects from their emitted thermal radiation. First demonstrated decades ago for objects much hotter than the air and recently extended to scenes with low temperature variations, this ranging has depended on reflected radiance being negligible. Downwelling radiance is especially problematic, sometimes causing large inaccuracies. In two new ranging methods, we use characteristic features from ozone absorption to estimate the contribution of reflected downwelling radiance. The quadspectral method gives a simple closed-form range estimate from four narrowband measurements, two at a water vapor absorption line and two at an ozone absorption line. The hyperspectral method uses a broader spectral range to improve accuracy while also providing estimates of temperature, emissivity profiles, and contributions of downwelling from a collection of zenith angles. Experimental results demonstrate improved ranging accuracy, in one case reducing error from over 100 m when reflected light is not modeled to 6.8 m with the quadspectral method and 1.2 m with the hyperspectral method.
Figures & tables
Fig. 1 : Effect of downwelling radiance in absorption-based ranging: ( 1(a) ) RGB photograph of the scene shown for reference but not used in any computations; ( 1(b) ) hyperspectral absorption-based ranging results when neglecting downwelling radiance; ( 1(c) ) hyperspectral absorption-based ranging results when accounting for downwelling radiance (discussed in Section V ); ( 1(d) ) hyperspectral measurements at reflective panel, grass area, and sky pixels (to improve clarity, the green curve uses a distinct vertical scale provided on the right side). Pixels plotted in ( 1(d) ) are shown with × ’s in ( 1(b) ) and ( 1(c) ). The ozone absorption band is shown in gray. In natural scenes, where temperature variations are minimal, reflection from the overhead sky can contribute significantly to the measurements and cause range overestimations depending on the reflectivity of the material. The central innovation of this paper is to use measurements in the ozone absorption band to infer the strengths of downwelling radiance contributions from a set of zenith angles. This helps to separate the reflected downwelling radiance and improve the absorption-based ranging.
Fig. 2 : Conceptual figure for radiative transfer model. White and red arrows represent contributions from object emission and air emission to the observed spectrum, respectively. The gray arrow denotes reflected thermal radiation from incident ambient thermal radiation from all angles of the hemisphere, assuming a diffuse reflection model. The ambient thermal radiation is decomposed into downwelling and ground-sourced components.
Fig. 3 : Ozone absorption band in LWIR. ( 3(a) ) Comparison of ground-level horizontal attenuation and vertical downwelling radiance computed using SpectralCalc [ 43 ] for the U.S. Standard Atmosphere [ 46 ] . ( 3(b) ) Comparison of water vapor (0.1 volume mixing ratio) and ozone ( 5×10−5 volume mixing ratio) attenuation profiles. All spectra are calculated with 40 nm Gaussian ISRF. Most of the absorption features at ground level are due to water vapor and are common to both ground-level attenuation and downwelling radiance. The absorption feature near 9.5 \SIUnitSymbolMicrom is due to ozone and is only observed in downwelling radiance. This is because ozone concentration at ground level is negligible, while the ozone layer at higher altitudes affects the downwelling radiance.
Fig. 4 : Downwelling radiances at various zenith angles. ( 4(a) ) Downwelling radiances calculated using SpectralCalc [ 43 ] for 10 zenith angles ranging from 0∘ to 90∘ , bottom to top. The ozone absorption band and one water vapor absorption band are highlighted with solid red and black vertical lines respectively. The dashed curves represent the nearby transparent bands to the absorption line with the same color encoding. ( 4(b) ) Correlation between ozone difference (x-axis) and water vapor difference (y-axis), where blue circles represent the 10 zenith angles in ( 4(a) ). Shown water vapor features are at 8.58 \SIUnitSymbolMicrom –8.64 \SIUnitSymbolMicrom and the ozone features are at 9.49 \SIUnitSymbolMicrom –9.57 \SIUnitSymbolMicrom . The resulting correlation coefficient is s=0.94 , which is used to compute the estimate b^ of the bias b by scaling the measured radiance difference at the ozone absorption band, as described in ( 22 ).
Fig. 5 : Experimental results for bispectral ranging: ( 5(a) ) difference of measurements around ozone absorption band; ( 5(b) ) baseline bispectral method; ( 5(c) ) quadspectral method; ( 5(d) ) lidar data approximately registered to hyperspectral LWIR sensor for ground truth; and ( 5(e) ) comparison of range profiles over the vertical line highlighted in red, with pixels indexed from top to bottom. Histograms of range estimates over 8×8 patches are shown in ( 5(f) ) for the front checkerboard target; ( 5(g) ) for the rear checkerboard target; and ( 5(h) ) for the tree; the corresponding locations are indicated in the lidar map. The black pixels in the depth map represent the pixels with invalid estimations or no lidar data. The baseline bispectral method is vulnerable to reflected downwelling radiation resulting in overestimations for reflective objects. Including two more absorption lines around the ozone band and following the quadspectral method gives much more reliable results.
Method
Front Checkerboard
Rear Checkerboard
Tree
Bispectral
150.73 m
±
19.89 m
174.26 m
±
3.05 m
59.83 m
±
20.14 m
Quadspectral
22.84 m
±
7.20 m
26.77 m
±
4.63 m
46.47 m
±
25.33 m
Ground truth proxy
31.01 m
46.51 m
57.84 m
TABLE I : Means and standard deviations of experimental quadspectral results on 8×8 patches; histograms shown in Fig. 5 .
Fig. 6 : Experimental results for hyperspectral ranging: ( 6(a) ) hyperspectral method neglecting reflected downwelling radiance; ( 6(b) ) hyperspectral method accounting for reflected downwelling radiance, without TV regularization; ( 6(c) ) hyperspectral method accounting for reflected downwelling radiance, with TV regularization; ( 6(d) ) lidar data registered to hyperspectral LWIR sensor for ground truth; and ( 6(e) ) comparison vertical range profiles highlighted with red dashed line, with pixels indexed from top to bottom. Histograms of range estimates over 8×8 patches are shown in ( 6(f) ) for the front checkerboard target; ( 6(g) ) for the rear checkerboard target; and ( 6(h) ) for the tree; the corresponding locations are indicated in the lidar map. The black pixels in the depth map represent the pixels with invalid estimations or no lidar data. Including downwelling radiation in the model significantly removes the overestimations around pixel indices 100 and 150.
Fig. 7 : Temperature and emissivity results for hyperspectral method: ( 7(a) ) and ( 7(b) ) shows the temperature estimates for neglecting and including downwelling radiance, respectively; ( 7(c) ) and ( 7(d) ) shows emissivity clusters neglecting and including downwelling radiance, respectively; and ( 7(e) ) and ( 7(f) ) shows the mean spectra for the clusters. K-means with 5 groups is used to generate the emissivity clusters.
Method
Front Checkerboard
Rear Checkerboard
Tree
without down. corr.
200.00 m
±
0.00 m
200.00 m
±
0.00 m
63.00 m
±
8.85 m
with down. corr.
28.86 m
±
5.38 m
39.83 m
±
2.01 m
54.27 m
±
7.40 m
with down. corr. (TV)
28.82 m
±
5.31 m
39.84 m
±
1.78 m
53.99 m
±
2.01 m
Ground truth proxy
31.01 m
46.51 m
57.84 m
TABLE II : Means and standard deviations of experimental hyperspectral results on 8×8 patches; histograms shown in Fig. 6 .
Fig. 8 : Normalized and projected solid angle estimates ( Ωi/π ) in the hyperspectral method. The values show the weight of downwelling radiances at various angles from 0∘ to 90∘ zenith angles. The combination of downwelling radiances with estimated weights forms the reflected radiance component in the model.
Fig. 9 : Model fits for 6 pixels accross the scene, ( 9(a) ) more reflective part of the checkerboard target, ( 9(b) ) less reflective part of the checkerboard target, ( 9(c) ) sky, ( 9(d) ) grass, ( 9(e) ) foreground tree, ( 9(f) ) background forest. In each plot the solid blue curve represents the measurements, the solid orange curve represents the model fit accounting for reflected downwelling radiance and dashed yellow curve represent the model fit ignoring downwelling radiance.
Fig. 10 : Hyperspectral absorption-based ranging results on additional scenes. Each column corresponds to a specific path and step from the dataset available in [ 51 ] . The first row shows the RGB images for visual reference. Results in the second row are obtained with downwelling neglected [ 18 ] , and results in the third row use the downwelling mitigation introduced here.
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 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.
Zhihe Chen, Chen Fan, Shuo Liu +4
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 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