CoChat
Sign Up
Cited by 35Robotics and Sensor-Based LocalizationInertial Sensor and NavigationSpacecraft Dynamics and Control

Lunar Terrain Relative Navigation Using a Convolutional Neural Network for Visual Crater Detection

Open access
Downes, Lena M., Steiner, Ted J., How, Jonathan P.·2020·2020 American Control Conference (ACC)
DOI 10.23919/acc45564.2020.9147595
Open this paper in CoChat — free accountOpen PDF

Abstract

Terrain relative navigation can improve the precision of a spacecraft's position estimate by detecting global features that act as supplementary measurements to correct for drift in the inertial navigation system. This paper presents a system that uses a convolutional neural network (CNN) and image processing methods to track the location of a simulated spacecraft with an extended Kalman filter (EKF). The CNN, called LunaNet, visually detects craters in the simulated camera frame and those detections are matched to known lunar craters in the region of the current estimated spacecraft position. These matched craters are treated as features that are tracked using the EKF. LunaNet enables more reliable position tracking over a simulated trajectory due to its greater robustness to changes in image brightness and more repeatable crater detections from frame to frame throughout a trajectory. LunaNet combined with an EKF produces a decrease of 60% in the average final position estimation error and a decrease of 25% in average final velocity estimation error compared to an EKF using an image processing-based crater detection method when tested on trajectories using images of standard brightness.

AI snapshot

Written from the paper’s full text.

TL;DR

This paper integrates LunaNet, a CNN-based lunar crater detector, with an extended Kalman filter for simulated lunar terrain relative navigation. It reports that LunaNet improves robustness to image brightness and yields lower final position and velocity estimation errors than an EKF using a traditional image-processing crater detector.

Key findings

  • LunaNet tracks craters more persistently: average detection length is 13 frames for LunaNet versus 5 for the trinary edge detector on standard imagery, 13 versus 6 on bright imagery, and 13 versus 7 on dark imagery.
  • LunaNet detects 18 craters on average over a trajectory on standard imagery, compared with 16 for the trinary edge detector; on bright imagery LunaNet detects 19 versus 29, but the paper says many trinary detections were false matches.
  • With standard brightness, the LunaNet-based EKF achieves final average estimation errors of 0.53 m and 0.09 m/s, versus 1.33 m and 0.12 m/s for the trinary edge detector, corresponding to a 60% decrease in average final position error and a 25% decrease in average final velocity error.
  • LunaNet is more robust to brightness changes: with a 30% brightness increase or decrease, it detected the same average number of craters over twenty trajectories, while the trinary edge detector degraded and produced more false crater matches.

Methods

  • LunaNet uses a U-net-style CNN initialized on DeepMoon weights and additionally trained for ten epochs on 800 LROC Wide Angle Camera intensity images at 100 m/pixel, with 256×256 orthographic crops. Detections are post-processed by thresholding at >90% certainty, erosion, contour filtering, and ellipse fitting with a maximum minor-to-major axis ratio of 15%.
  • Detected craters are matched to known craters from a 5–20 km database and a >20 km database using expected craters from the EKF position estimate, least mean squares pairing, and affine-transform RANSAC to reject outlier matches.
  • A feature-based EKF estimates camera velocity, camera position, and feature positions in lunar-centered lunar-fixed coordinates. Simulations use true acceleration plus random noise on the order of 0.1 m/s² and a 2.5 s time step; position and velocity errors are evaluated over 100 trajectories, and brightness comparisons use Monte Carlo simulations over 20 trajectories.

Limitations

  • The paper does not include an explicit limitations section.
  • Evident from the text: evaluation is simulation-only, using simulated camera imagery and orbital trajectories rather than real lunar flight data or real camera hardware.
  • Evident from the text: the EKF is initialized on true position and velocity, and brightness robustness is tested only for ±30% changes from the training/tuning set.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.23919/acc45564.2020.9147595

Cites

26 of 26 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Claims the paper makes

0 of 17 cited sources checked · 0 supported · 0 flagged

  • Findingown data

    LunaNet is less sensitive to changes in brightness levels in the image.

    “Another key improvement of LunaNet is that it is less sensitive to changes in brightness levels in the image.”
    Quote in the paperSubstantial claim
    • Rests on an assumption: A 30% difference in brightness level could unknowingly occur due to differences in time of day, exposure of the camera, or albedo properties of the local lunar surface.
  • Findingown data

    LunaNet tracks craters on average for twice as many frames as the trinary edge detector does.

    “LunaNet tracks craters on average for twice as many frames as the trinary edge detector does.”
    Quote in the paperSubstantial claim
  • Findingreasoning

    The inconsistency in feature detection of the trinary edge detector results in more false matches, where detected craters are mistakenly matched to the wrong database craters.

    “The inconsistency in feature detection of the trinary edge detector also results in more false matches, where detected craters are mistakenly matched to the wrong database craters.”
    Quote in the paperSubstantial claim
  • Findingreasoning

    The trinary edge detector has many trajectories with estimation errors that spike, diverge and drift for large portions of the simulation.

    “In contrast, the trinary edge detector in Fig. 2(a) has many trajectories with estimation errors that spike, diverge and drift for large portions of the simulation.”
    Quote in the paperSubstantial claim
    • The authors state this as an assumption: The motion of the spacecraft can be considered planar for short segments of orbits.
  • Findingreasoning

    LunaNet achieves more consistently long feature tracks, meaning that once a crater is initialized, it is repeatedly detected in subsequent frames more times than the trinary edge detector does.

    “LunaNet achieves more consistently long feature tracks, meaning that once a crater is initialized, it is repeatedly detected in subsequent frames more times than the trinary edge detector does.”
    Quote in the paperSubstantial claim
  • May combine two claims. 30
  • The quote only partly states this claim. 2
  • Classified as probably not a substantial claim. 5

  • May combine two claims.

    Findingown data

    LunaNet combined with an EKF produces a decrease of 60% in the average final position estimation error and a decrease of 25% in average final velocity estimation error compared to an EKF using an image processing-based crater detection method when tested on trajectories using images of standard brightness.

    “LunaNet combined with an EKF produces a decrease of 60% in the average final position estimation error and a decrease of 25% in average final velocity estimation error compared to an EKF using an image processing-based crater detection method when tested on trajectories using images of standard brightness.”
    • Also stated in: § VII. CONCLUSIONS · plain
      “A LunaNetbased EKF obtains a decrease of 60% in the average final position estimation error and a decrease of 25% in the average final velocity estimation error for standard brightness images when compared to an EKF based on the trinary edge detector.”
    Quote in the paperSubstantial claim
    • The authors state this as an assumption: The motion of the spacecraft can be considered planar for short segments of orbits.

  • May combine two claims.

    Findingown data

    The final average estimation error of the trinary edge detector is 1.33 m and 0.12 m/s.

    “The final average estimation error of the trinary edge detector is 1.33 m and 0.12 m/s.”
    Quote in the paperSubstantial claim
    • The authors state this as an assumption: The motion of the spacecraft can be considered planar for short segments of orbits.

  • May combine two claims.

    Findingown data

    The trinary edge detector is highly tuned to images with specific brightness levels, while LunaNet does not require additional training to perform comparably well on images with different levels of brightness.

    “The trinary edge detector is highly tuned to images with specific brightness levels, while LunaNet does not require additional training to perform comparably well on images with different levels of brightness.”
    Quote in the paperSubstantial claim
    • Rests on an assumption: A 30% difference in brightness level could unknowingly occur due to differences in time of day, exposure of the camera, or albedo properties of the local lunar surface.

  • May combine two claims.

    Findingown data

    The EKF using LunaNet experiences fewer spikes in estimation error and converges more quickly than the EKF using the trinary edge detector on standard brightness images.

    “The EKF using LunaNet experiences fewer spikes in estimation error and converges more quickly than the EKF using the trinary edge detector in Fig. 2(a), which shows the norm of the position estimation error over the same 100 trajectories.”
    Quote in the paperSubstantial claim
    • The authors state this as an assumption: The motion of the spacecraft can be considered planar for short segments of orbits.

  • May combine two claims.

    Findingown data

    The trinary edge detector struggles to persistently detect the same craters in subsequent frames, contributing to shorter length of crater detection tracks in each trajectory.

    “The trinary edge detector struggles to persistently detect the same craters in subsequent frames.This contributes to the shorter length of crater detection tracks in each trajectory for the trinary edge detector, as characterized in Table I.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Overall, LunaNet results in fewer estimation error spikes and faster convergence on images of different brightness levels when compared to the performance of the trinary edge detector on the same images and trajectories.

    “Overall, LunaNet results in fewer estimation error spikes and faster convergence on images of different brightness levels when compared to the performance of the trinary edge detector on the same images and trajectories.”
    Quote in the paperSubstantial claim
    • Rests on an assumption: A 30% difference in brightness level could unknowingly occur due to differences in time of day, exposure of the camera, or albedo properties of the local lunar surface.

  • May combine two claims.

    Findingown data

    On the brighter image, LunaNet detects all but one of the same craters as in the standard image, while the trinary edge detector detects only one of the same craters and produces three false matches.

    “On the brighter image, LunaNet detects all but one of the same craters as in the standard image.In contrast, the trinary edge detector detects only one of the same craters as in the standard image.It also produces three false matches, marked in red.”
    Quote in the paperSubstantial claim
    • Rests on an assumption: A 30% difference in brightness level could unknowingly occur due to differences in time of day, exposure of the camera, or albedo properties of the local lunar surface.

  • May combine two claims.

    Findingreasoning

    The trinary edge detector struggles to repeatedly detect craters in subsequent frames after initialization, while LunaNet persistently detects craters from frame to frame.

    “The trinary edge detector struggles to repeatedly detect craters in subsequent frames after initialization, while LunaNet persistently detects craters from frame to frame.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingreasoning

    The trinary edge detector experienced a degradation in performance and detected a higher number of false crater matches.

    “In contrast, the trinary edge detector experienced a degradation in performance and detected a higher number of false crater matches.”
    Quote in the paperSubstantial claim
    • Rests on an assumption: A 30% difference in brightness level could unknowingly occur due to differences in time of day, exposure of the camera, or albedo properties of the local lunar surface.

  • May combine two claims.

    Findingown data

    LunaNet detects almost all of the same craters in both standard and 30% brighter images, and human inspection revealed all of them to be true matches.

    “LunaNet detects almost all of the same craters in both images, and human inspection revealed all of them to be true matches.”
    Quote in the paperSubstantial claim
    • Rests on an assumption: A 30% difference in brightness level could unknowingly occur due to differences in time of day, exposure of the camera, or albedo properties of the local lunar surface.

  • May combine two claims.

    Findingreasoning

    A LunaNet-based EKF obtains higher accuracy and converges more quickly than an EKF based on a traditional crater detector.

    “A LunaNet-based EKF obtains higher accuracy and converges more quickly than an EKF based on a traditional crater detector.”
    Quote in the paperSubstantial claim
    • The authors state this as an assumption: The motion of the spacecraft can be considered planar for short segments of orbits.

  • May combine two claims.

    Findingown data

    Although the trinary edge detector detects a higher number of craters with changes in image brightness, a greater percentage of these detections were incorrectly matched to the wrong known craters.

    “Although the trinary edge detector detects a higher number of craters with changes in image brightness, visual inspection revealed that a greater percentage of these detections were incorrectly matched to the wrong known craters.”
    Quote in the paperSubstantial claim
    • Rests on an assumption: A 30% difference in brightness level could unknowingly occur due to differences in time of day, exposure of the camera, or albedo properties of the local lunar surface.

  • May combine two claims.

    Findingown data

    LunaNet converges faster and experiences a smaller degradation in performance than the trinary edge detector, even on images that are brighter or darker than those that it was trained on.

    “LunaNet converges faster and experiences a smaller degradation in performance than the trinary edge detector, even on images that are brighter or darker than those that it was trained on.”
    Quote in the paperSubstantial claim
    • Rests on an assumption: A 30% difference in brightness level could unknowingly occur due to differences in time of day, exposure of the camera, or albedo properties of the local lunar surface.

  • May combine two claims. The quote only partly states this claim.

    Findingown data

    The trinary edge detector detects completely different craters on the brighter image than it did in the standard image, and only one of its accepted and matched detections was properly matched to a known crater.

    “The trinary edge detector detects completely different craters than it did in the standard image, and human inspection revealed that only one of the accepted and matched detections was properly matched to a known crater.”
    Quote in the paperSubstantial claim
    • Rests on an assumption: A 30% difference in brightness level could unknowingly occur due to differences in time of day, exposure of the camera, or albedo properties of the local lunar surface.

  • May combine two claims.

    Findingown data

    LunaNet detects 18 craters on average over a trajectory, where the trinary edge detector detects 16 craters on average.

    “Table II shows numerical measures of how LunaNet outperforms the trinary edge detector in average number of craters detected, with LunaNet detecting 18 craters on average over a trajectory, where the trinary edge detector detects 16 craters on average.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    The final average estimation errors using LunaNet are 0.53 m and 0.09 m/s.

    “The final average estimation errors using LunaNet are 0.53 m and 0.09 m/s.”
    Quote in the paperSubstantial claim
    • The authors state this as an assumption: The motion of the spacecraft can be considered planar for short segments of orbits.

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation

    CraterIDNet applied a fully convolutional network (FCN) to both visually detect craters in images and match those detected craters to known craters.

    “CraterIDNet [8] applied a fully convolutional network (FCN) [9] to both visually detect craters in images and match those detected craters to known craters.”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites [8] source

      Not checked in the free run.

    • cites [9] source

      Not checked in the free run.

  • May combine two claims.

    Interpretationreasoning

    LunaNet's repeatable crater detections from frame to frame across a trajectory and increased robustness to changes in camera lighting conditions enable an EKF with increased precision position and velocity estimation and more reliable performance in variable lighting conditions.

    “LunaNet enables more reliable position tracking over a simulated trajectory due to its greater robustness to changes in image brightness and more repeatable crater detections from frame to frame throughout a trajectory.”
    • Also stated in: § I. INTRODUCTION · plain
      “LunaNet's repeatable detections from frame to frame across a trajectory and increased robustness to changes in camera lighting conditions enable an EKF that has increased precision position and velocity estimation with more reliable performance in variable lighting conditions.”
    Quote in the paperSubstantial claim
    • Rests on an assumption: A 30% difference in brightness level could unknowingly occur due to differences in time of day, exposure of the camera, or albedo properties of the local lunar surface.

  • May combine two claims.

    Premisecitation

    The authors have previously developed LunaNet, a neural network-based crater detector that visually detects craters in a camera image and matches those detected craters to a database of known lunar craters with absolute latitudes and longitudes.

    “We have previously developed a system that visually detects craters in a camera image using a neural network and matches those detected craters to a database of known lunar craters with absolute latitudes and longitudes [5].This crater detector is called LunaNet.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [5] source

      Not checked in the free run.

  • May combine two claims.

    Premisecitation

    Many crater locations have already been catalogued since craters are present on nearly the entire lunar surface.

    “Many crater locations have already been catalogued since craters are present on nearly the entire lunar surface [7].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [7]

      Not checked in the free run.

  • May combine two claims.

    Premisecitation

    Previous crater detectors based on traditional image processing methodologies like thresholding, edge detection, and filtering are highly sensitive to lighting conditions, noise, camera parameters, and viewing angles, and require precise modeling and specific tuning for different sets of images.

    “Previous works on lunar TRN systems have developed crater detectors that are based upon traditional image processing methodologies like thresholding, edge detection, and filtering [3].These crater detectors are highly sensitive to lighting conditions, noise, camera parameters, and viewing angles, and require precise modeling and specific tuning for different sets of images.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [3] source

      Not checked in the free run.

  • May combine two claims.

    Premisecitation

    Past lunar landing errors have been as high as several kilometers because inertial navigation systems drift and accumulate error over time.

    “As such, past lunar landing errors have been as high as several kilometers [1] because inertial navigation systems drift and accumulate error over time.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [1]

      Not checked in the free run.

  • Classified as probably not a substantial claim. May combine two claims.

    Premisecitation

    Neural networks have been successfully applied as a robust solution to many computer vision problems due to their strength in generalizing from one dataset to another, but have rarely been used in space applications due to a perceived lack of confidence in their measurements.

    “Neural networks have been successfully applied as a robust solution to many computer vision problems due to their strength in generalizing from one dataset to another, but have rarely been used in space applications due to a perceived lack of confidence in their measurements [4].”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [4] source

      Not checked in the free run.

  • May combine two claims.

    Premisecitation

    [3]'s crater identification method was successful if the position information provided was very close to the true position, but resulted in a large number of mismatches otherwise.

    “This approach was successful if the position information provided was very close to the true position, but resulted in a large number of mismatches otherwise.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [3] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    DeepMoon applied a CNN to detect craters from a digital elevation map (DEM) represented as overhead imagery.

    “Another recent system, DeepMoon [10], applied a CNN to detect craters from a digital elevation map (DEM) represented as overhead imagery.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [10] source

      Not checked in the free run.

  • Classified as probably not a substantial claim.

    Premisecitation

    The system of [3] used a crater detector based on traditional image processing methods, referred to as the trinary edge detector due to its method of finding craters through trinary thresholding and edge detection.

    “This lunar terrain relative navigation system used a crater detector based on traditional image processing methods, which we refer to as the trinary edge detector due to its method of finding craters through trinary thresholding and edge detection.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [3] source

      Not checked in the free run.

  • May combine two claims. The quote only partly states this claim.

    Interpretationreasoning

    More consistently long feature tracks cause estimation uncertainty to decrease more rapidly, resulting in a more reliable position estimate.

    “This quality causes estimation uncertainty to decrease more rapidly, resulting in a more reliable position estimate.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Premisecitation

    [3] demonstrated the potential for craters to improve the state estimation of a spacecraft in lunar orbit, but struggled to consistently detect the same craters from frame to frame in a trajectory and experienced significant sensitivity to image quality, brightness, and shadowing.

    “This work demonstrated the potential for craters to improve the state estimation of a spacecraft in lunar orbit.However, it struggled to consistently detect the same craters from frame to frame in a trajectory and experienced significant sensitivity to image quality, brightness, and shadowing.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [3] source

      Not checked in the free run.

  • May combine two claims.

    Interpretationreasoning

    Higher numbers of false matches cause the EKF to diverge and result in less reliable state estimation.

    “These higher numbers of false matches cause the EKF to diverge and result in less reliable state estimation.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationreasoning

    The trinary edge detector's estimation error indicates that this method is less reliable and more prone to false detections, false matches, and overall lower numbers of detections.

    “The trinary edge detector's estimation error indicates that this method is less reliable and more prone to false detections, false matches, and overall lower numbers of detections.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation

    [12] incorporated random sample consensus (RANSAC) to check whether the translation vectors between detected craters and known craters were consistent in images where multiple craters were detected; if a crater match produced a translation vector that was an outlier, then that match was rejected.

    “This approach incorporated random sample consensus (RANSAC) [13] to check whether the translation vectors between detected craters and known craters were consistent in images where multiple craters were detected.If a crater match produced a translation vector that was an outlier from the other crater matches, then that match was rejected.”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites [12] source

      Not checked in the free run.

    • cites [13] source

      Not checked in the free run.

  • May combine two claims.

    Interpretationreasoning

    False matches are usually thrown out by RANSAC in the crater matching system, but occasionally they are accepted as good matches, contributing to shorter average tracking length and causing the EKF to diverge.

    “These false matches are usually thrown out by RANSAC in the crater matching system, but occasionally they are accepted as good matches.These occurrences contribute to the shorter average tracking length, since false matches could be matched to different database craters in consecutive frames, thereby causing the EKF to diverge.”
    Quote in the paperSubstantial claim
32 of 32

  • The paper relies on this without support.

    Many locations of interest for future lunar landing missions are surrounded by or close to hazardous terrain, motivating the requirement for increased estimation precision throughout the mission.

    “However, many of the locations of interest for future lunar landing missions are surrounded by or are close to hazardous terrain, motivating the requirement for increased estimation precision throughout the mission.”

  • The paper relies on this without support.

    A detected crater's location can be uniquely identified based on the local constellation of nearby detected craters.

    “In addition, a detected crater's location can be uniquely identified based on the local constellation of nearby detected craters.”

  • The paper relies on this without support.

    True craters are generally circular, and detections that are more elliptical tend to be false detections.

    “The ellipses are not allowed to be more elliptical because true craters are generally circular, and detections that are more elliptical tend to be false detections.”

  • The authors state this as an assumption.

    The motion of the spacecraft can be considered planar for short segments of orbits.

    “The state propagation equations are linear approximations in the following equations since the motion can be considered planar for short segments of orbits,”

  • The paper relies on this without support.

    A 30% difference in brightness level could unknowingly occur due to differences in time of day, exposure of the camera, or albedo properties of the local lunar surface.

    “A 30% difference in brightness level could unknowingly occur due to differences in time of day, exposure of the camera, or albedo properties of the local lunar surface.”
5 of 5

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

  • Feature matching algorithms that localize craters and other landmarks against known databases based only on optical inputs experience extreme sensitivity to changes in illumination conditions, camera angle, and other numerical errors.

    “Feature matching algorithms that localize craters and other landmarks against known databases based only on optical inputs i.e., images experience extreme sensitivity to changes in illumination conditions, camera angle, and other numerical errors14,15.”

    From Probabilistic regression for autonomous terrain relative navigation via multi-modal feature learning (2024) · cites this paper as [14]

    The Introduction reports that the traditional crater detector is sensitive to image quality, brightness, and shadowing, and that inaccurate position information can cause crater mismatches. The Simulation Results show false matches and estimation errors when image brightness changes. The paper does not establish sensitivity to camera angle or to numerical errors generally, and its matching system also uses a spacecraft position estimate rather than optical inputs alone. (LLM, conf=0.86)

    Experimental — accuracy not yet measured

  • The ODN crater detections are robust to disturbances or modifications of image quality due to Gaussian noise, shot noise, brightness levels, and different illumination conditions.

    “the amount of crater detections that ODN-detector produces is robust to disturbances or modifications of the image quality due to Gaussian noise, shot noise, brightness levels, as well as different illumination conditions.”

    From Optical navigation for Lunar landing based on Convolutional Neural Network crater detector (2022) · cites this paper as [22]

    Section VI (page unavailable) reports that LunaNet’s crater detections remain relatively consistent when images are 30% brighter or darker. The paper does not identify its detector as ODN or report tests of Gaussian noise, shot noise, or distinct illumination conditions; Section III mentions robustness to noise only generally. (LLM, conf=0.84)

    Experimental — accuracy not yet measured

  • State-of-the-art crater detectors make use of Object Detection Networks and image segmentation.

    “State of the art detectors make use of Object Detection Networks (ODN) [18,19] image segmentation [20,21,22].”

    From Optical navigation for Lunar landing based on Convolutional Neural Network crater detector (2022) · cites this paper as [22]

  • State-of-the-art crater detection networks rely heavily on the U-Net architecture to identify landmarks by means of image segmentation.

    “State of the art crater detection networks rely heavily on the so called U-Net architecture [34,35] to identify landmarks by means of image segmentation [20,21,22].”

    From Optical navigation for Lunar landing based on Convolutional Neural Network crater detector (2022) · cites this paper as [22]

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Each contradiction between two passages listed here was confirmed by an independent check.

  • Ellipse criterion excludes circular craters

    Minor

    A circle has a minor-to-major axis ratio of 100%, so a maximum ratio of 15% would exclude circular detections. This conflicts with the stated aim of retaining generally circular craters and rejecting more elliptical detections.

  • Crater averages differ across brightness levels

    Minor

    The conclusion says LunaNet detected the same average number of craters after either brightness change, but Table II reports 18 for standard imagery, 19 for bright imagery, and 16 for dark imagery.

2 of 2

Cites

Flagged first, then by citations

Cited by

Sorted by citations

Related

Authors
Downes, L. M., Steiner, T. J., & How, J. P.
Affiliations
  • Massachusetts Institute of Technology (US)
  • Draper Laboratory (US)
ORCID
Venue
2020 American Control Conference (ACC)
Publisher
IEEE
Published
2020-07
Pages
4448-4453
ISSN
0743-1619 (ISSN-L)
DOI
10.23919/acc45564.2020.9147595
OpenAlex
W3042503233
Cited by
35
References
23
Citations by year
2026: 2 · 2025: 8 · 2024: 9 · 2023: 6 · 2022: 6 · 2021: 3 · 2020: 1
FWCI
5.89
Type
conference paper
Language
en
Open access
Locations
  • Proceedings of the ... American Control Conference/Proceedings of the American Control Conference (publishedVersion)
  • arXiv (Cornell University) (submittedVersion)
  • DSpace@MIT (Massachusetts Institute of Technology) (submittedVersion, cc-by-nc-sa)
  • arXiv (Cornell University)
Funders
Amazon Web Services
Topics
  • Physical Sciences › Engineering › Aerospace Engineering › Robotics and Sensor-Based Localization
  • Physical Sciences › Engineering › Aerospace Engineering › Inertial Sensor and Navigation
  • Physical Sciences › Engineering › Aerospace Engineering › Spacecraft Dynamics and Control
Keywords
convolutional neural networks, extended Kalman filter, lunar craters, spacecraft trajectory simulation

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Claims read by deepseek-v4.1-flash under definition version 3 (extract-v3) on 2026-10-10.

Classified by jev-1.13-20260917 (gold-guideline-v0.2) on 2026-10-10.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review