Student or educator? Sign up with your school email and get 6 months of Pro free.
CoChat

The DOI matches its Crossref record; 96 of the 105 references it cites were found in a registry or an index; 10 of the 73 sources its claims cite were checked for support.

Semantic Terrain Segmentation in the Navigation Vision of Planetary Rovers—A Systematic Literature Review

Open access

Abstract

This systematic literature review (SLR) identifies key developments, available datasets, and future challenges for semantic terrain segmentation in the navigation vision of planetary rovers. It covers 30 primary papers spanning navigation, geological analysis, and exploration.

Overview

CoChat Summary

TL;DR

This systematic literature review analyzes 30 studies on semantic terrain segmentation in planetary rover navigation vision, covering methods, data, targets, and evaluation metrics. It finds growing deep-learning activity but no existing solution that simultaneously satisfies pixel-level segmentation, real-time inference, and onboard hardware, nor an open, pixel-level annotated, real-world image dataset.

Key findings

  • The review included 30 studies. Application areas were navigation (16 studies), geological analysis (7), exploration efficiency (10), and others (3), with overlaps. Rock was a dominant target: 54% of studied targets were rocks, and half of the Included Study addressed rock, yet no on-time rock segmentation solution with high pixel-level accuracy on onboard hardware exists yet.
  • Data availability is limited: 60% of data used were open-source or had open source data, but 53% of open-source datasets had fewer than 1000 images. Most past space exploration data are raw and unannotated. The review found three prior open datasets and four new datasets.
  • Sensor and format distributions: 80% of monocular images were used versus 20% stereo image pairs; 63% of studies used grayscale images and only 3% used depth images. Fifty percent of the Included Study did not provide the experimental setting.
  • Method trends and performance: deep learning-based studies rapidly increased since 2018. Accuracy and IoU were the most used criteria; reference [67] achieved 99.58% accuracy with modified U-Net++, and reference [66] achieved IoU for the sky region of 0.9066. Unsupervised learning was used in 26% of studies, with precision in reference [59] only about 65% versus more than 99% in reference [65].

Methods

  • A systematic literature review was conducted following guideline references [33,34], using PICOC criteria. Searches covered IEEE Xplore, Web of Science, and Scopus, completed in May 2022, with four parallel search conditions using “AND”: terrain* OR *rock* OR sky*; segment*; planetary OR mars OR lunar OR Martian OR moon; and image* OR vision OR visual.
  • The identification phase found 320 candidate studies: 73 from IEEE Xplore, 64 from Web of Science, and 183 from Scopus. Screening applied language, duplication/document type, abstract/title, full-text, and quality assessment criteria; 30 studies were included for analysis.
  • Quality assessment used five elements (theory robustness, implication for practice, methodology/data supporting arguments, generalizability, and contribution) with five levels (“0” Absence, “1” Low, “2” Medium, “3” High, “Not applicable”). Data extraction covered bibliographic details, targets, data format, data source, image number, sensor/camera type, method classification, solution details, metrics, and qualitative results.

Limitations

  • The paper states three limitations: the manual identification step is recommended for software technology evaluation; candidate studies were identified by a single researcher while research questions, search strategy, exclusion criteria, and quality assessment were reviewed by other researchers; and the definitions of “Classical Image Processing-based Study,” “Machine Learning-based Study,” and “Deep Learning-based Study” are proposed only in this study and do not belong to common practice.
  • The paper also states that the search field for different keywords might contain bias because IEEE Xplore, Web of Science, and Scopus used different search fields (“All Metadata,” “AB (abstract),” and “TITLE-ABS-KEY”), decided by a single researcher and reviewed by another; and that image processing, machine learning, and deep learning are not independent topics in common practice, so the categories might not be precisely divided.
  • The text makes evident that the review is limited to English-written major international journals and conferences, so relevant studies in national journals/conferences or non-English languages might not be identified; the search was completed in May 2022, so later studies are outside its scope.

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

Integrity signals

Registry

Crossref · 10.3390/s22218393

Cites

96 of 96 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Figures

Figure 1 The flow diagram of this systematic literature review. The search process has been separated into three phases, “Identification phase”, “Eligibility phase”, and “Inclusion phase”. The screening follows the inclusion and exclusion criteria depicts in Section 2.3. The “n” in each frame refers to the number of studies for the corresponding action or category.
Figure 1 The flow diagram of this systematic literature review. The search process has been separated into three phases, “Identification phase”, “Eligibility phase”, and “Inclusion phase”. The screening follows the inclusion and exclusion criteria depicts in Section 2.3. The “n” in each frame refers to the number of studies for the corresponding action or category.CC BY · PubMed Central
Figure 2 The statistics of the new studies in the Included Study. (a,b) Refer to the increasing and cumulative number of the Included Study.
Figure 2 The statistics of the new studies in the Included Study. (a,b) Refer to the increasing and cumulative number of the Included Study.CC BY · PubMed Central
Figure 3 The distribution by study type in the Included Study. The “conference findings” and “journal achievements” refer to the studies published in the conferences and journals, respectively.
Figure 3 The distribution by study type in the Included Study. The “conference findings” and “journal achievements” refer to the studies published in the conferences and journals, respectively.CC BY · PubMed Central
Figure 4 The distribution by publisher in the Included Study.
Figure 4 The distribution by publisher in the Included Study.CC BY · PubMed Central
Figure 5 The distribution by the experimental setting.
Figure 5 The distribution by the experimental setting.CC BY · PubMed Central
Figure 6 The ratio distribution by the target in the Included Study.
Figure 6 The ratio distribution by the target in the Included Study.CC BY · PubMed Central
Figure 7 The distribution by camera model and image format in the Included Study. (a) Refers to the ratio of the monocular camera model in the Included Study. (b) Refers to the distribution by image format in the Included Study.
Figure 7 The distribution by camera model and image format in the Included Study. (a) Refers to the ratio of the monocular camera model in the Included Study. (b) Refers to the distribution by image format in the Included Study.CC BY · PubMed Central
Figure 8 The classification rules of the proposed three categories for the included study (the classical image processing-based study, the machine learning-based study, and the deep learning-based study).
Figure 8 The classification rules of the proposed three categories for the included study (the classical image processing-based study, the machine learning-based study, and the deep learning-based study).CC BY · PubMed Central
Figure 9 The increasing trends of the Included Study by method categories (the classical image processing-based studies, machine learning-based studies, and deep learning-based studies).
Figure 9 The increasing trends of the Included Study by method categories (the classical image processing-based studies, machine learning-based studies, and deep learning-based studies).CC BY · PubMed Central
Figure 10 The distributions by dataset types and the number of images in the Included Study. (a) Refers to the distribution by dataset type, and (b) refers to the distribution by the number of images in the source data.
Figure 10 The distributions by dataset types and the number of images in the Included Study. (a) Refers to the distribution by dataset type, and (b) refers to the distribution by the number of images in the source data.CC BY · PubMed Central
Figure 11 The number of open datasets in the Included Study. The “NASA”, “Katwijk”, and “Devon” refer to the NASA image set of Mars Exploration Rovers, the Katwijk beach planetary rover dataset, and the Devon Island rover navigation dataset, respectively.
Figure 11 The number of open datasets in the Included Study. The “NASA”, “Katwijk”, and “Devon” refer to the NASA image set of Mars Exploration Rovers, the Katwijk beach planetary rover dataset, and the Devon Island rover navigation dataset, respectively.CC BY · PubMed Central

Verification

Claims the paper makes

8 Valid · 12 Not valid · 34 Not checked · 93 Own results and interpretations

43 more claims

Create a free account to see all 147 claims and to verify the 10 references still open.

See all claims — free account
  • Probably not a substantial claim. 58
  • It may state two separate claims as one. 67
  • Borderline: one of the tests for a substantial claim is unclear. 1
  • The quote only partly states this claim. 3

Create a free account to see these claims.

Self-check

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

1 of 1

Citations

Cites

Research works printed but not linked: 4

  1. IEEE (2021, November 16). IEEE Xplore. Available online: https://ieeexplore.ieee.org/Xplore/home.jsp.
  2. (2021, November 16). Elsevier Scopus. Available online: https://www.scopus.com/search/form.uri?display=basic#basic.
  3. Song, Y., and Shan, J. (2006, January 1–5). A Framework for Automated Rock Segmentation from the Mars Exploration Rover Imagery. Proceedings of the ASPRS 2006 Annual Conference, Reno, NV, USA.
  4. Reid, 2018, Machine Learning Applications for Safe and Efficient Rover Mobility Operations and Planning, Proceedings of the 69th International Astronautical Congress, Volume 2, 47243

Cited by

See all 15

Metadata

Authors
Kuang, B., Gu, C., Rana, Z. A., Zhao, Y., Sun, S., & Nnabuife, S. G.
Affiliations
  • Cranfield University (GB)
  • Civil Aviation University of China (CN)
Corresponding author
Sun, Shuang
Venue
Sensors
Publisher
MDPI AG
Published
2022-11-01
Volume
22
Issue
21
Pages
8393
ISSN
1424-8220 (ISSN-L)
OpenAlex
W4308126638
Cited by
15
References
105
Citations by year
2026: 4 · 2025: 3 · 2024: 5 · 2023: 3
FWCI
0.23
Type
journal article
Language
en
Open access
Locations
  • Sensors (publishedVersion, cc-by)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
  • CERES (Cranfield University) (submittedVersion, cc-by)
Funders
  • Scientific Research Program of the Tianjin Education Commission (2019KJ120, 3122015C006)
  • Basic Scientific Research of Central Universities (2019KJ120, 3122015C006)
Topics
  • Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Advanced Neural Network Applications
  • Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Robotics and Sensor-Based Localization
  • Physical Sciences › Engineering › Civil and Structural Engineering › Infrastructure Maintenance and Monitoring
Keywords
planetary rovers, pixel-level segmentation, real-time inference, annotated dataset, geological analysis, rover autonomy
MeSH terms
Artificial Intelligence, Data Collection, Publications, Semantics

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.

The claim verdicts this report counts are experimental: their accuracy is not yet measured.

A language model drew the claims from the paper on 2026-10-10, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-10-10 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

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

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.