Monitoring road infrastructures from satellite images in Greater Maputo

Arianna Burzacchi; Matteo Landr\`o; Simone Vantini

arxiv: 2409.06406 · v2 · submitted 2024-09-10 · 📊 stat.AP

Monitoring road infrastructures from satellite images in Greater Maputo

Arianna Burzacchi , Matteo Landr\`o , Simone Vantini This is my paper

Pith reviewed 2026-05-23 20:59 UTC · model grok-4.3

classification 📊 stat.AP

keywords satellite imageryroad pavement classificationpaved versus unpavedRGB pixel distributionobject-oriented analysisMaputoinfrastructure monitoringdeveloping countries

0 comments

The pith

An object-oriented analysis of RGB pixel distributions in satellite images classifies roads as paved or unpaved.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper develops an automatic pipeline that uses satellite imagery to label road segments as paved or unpaved in regions where such data is missing from official databases. It extracts each road as a distinct object and classifies it according to the statistical distribution of RGB values among its pixels. The authors show that this yields accurate results at low cost and can be repeated in other cities. A reader would care because surface-type information is a basic input for designing workable mobility systems in developing areas.

Core claim

The paper claims that an object-oriented approach, in which each road segment is classified by the distribution of its pixels in RGB space, produces accurate paved-versus-unpaved labels from satellite images and does so inexpensively and in a form that transfers readily to other cities.

What carries the argument

The object-oriented classification that assigns each extracted road segment a label on the basis of the distribution of its RGB pixel values.

If this is right

Road-network databases in developing countries can receive pavement-type attributes without costly ground surveys.
Mobility-system planning can incorporate surface condition as a design variable.
The same satellite-based pipeline can be applied directly to other cities that have comparable imagery.
Data gaps on basic infrastructure attributes can be reduced at scale.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

The resulting maps could be used to prioritize maintenance spending toward unpaved segments.
Overlaying the classified roads with traffic or population layers might reveal equity or efficiency patterns not examined in the paper.
Extension to multi-spectral bands or time-series imagery could reduce residual errors from single-date RGB data.

Load-bearing premise

The RGB pixel-value distribution inside each road object extracted from the imagery is enough by itself to separate paved from unpaved surfaces.

What would settle it

A field check or independent database comparison in Greater Maputo that finds frequent misclassifications traceable to shadows, vegetation, or image resolution rather than surface type.

Figures

Figures reproduced from arXiv: 2409.06406 by Arianna Burzacchi, Matteo Landr\`o, Simone Vantini.

**Figure 2.** Figure 2: Cloud of points in the RGB space (on the left) and its projection in [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗

**Figure 3.** Figure 3: Cloud of points in the RGB space of a specific road satellite image. [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗

**Figure 4.** Figure 4: Maputo road network colored by known pavement type (top images) [PITH_FULL_IMAGE:figures/full_fig_p013_4.png] view at source ↗

**Figure 5.** Figure 5: Distribution of paved and unpaved labels among street types [PITH_FULL_IMAGE:figures/full_fig_p017_5.png] view at source ↗

read the original abstract

The information about pavement surface type is rarely available in road network databases of developing countries although it represents a cornerstone of the design of efficient mobility systems. This research develops an automatic classification pipeline for road pavement which makes use of satellite images to recognize road segments as paved or unpaved. The proposed methodology is based on an object-oriented approach, so that each road is classified by looking at the distribution of its pixels in the RGB space. The proposed approach is proven to be accurate, inexpensive, and readily replicable in other cities.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

The paper applies standard object-based RGB pixel distribution classification to paved/unpaved roads in Greater Maputo but supplies no accuracy numbers or robustness checks, so the central claim cannot be evaluated.

read the letter

The main takeaway is that this work takes an established object-oriented satellite image method—classifying each road segment by the RGB color spread of its pixels—and applies it to Greater Maputo to label roads as paved or unpaved. That is the extent of what is new here: a location-specific use case rather than a new technique or theoretical result. The approach is simple and avoids fitted models or extra parameters, which keeps it cheap and potentially easy to copy in other cities with similar data gaps on road surfaces. That practical focus on a real infrastructure problem in a developing-country setting is the part that holds up without overclaiming. The authors correctly note that pavement type information is often missing from local databases and that satellite data could help fill it at low cost. On the evidence side, the abstract states the method is proven accurate, inexpensive, and replicable, but it gives no error rates, no ground-truth comparison, no sample sizes, and no mention of how common satellite issues like shadows, haze, vegetation overlap, or illumination changes are handled. The stress-test concern about overlapping RGB distributions under real conditions therefore lands directly on the presented material. Without those details the accuracy assertion stays untestable. This is aimed at practitioners doing urban planning or road maintenance in data-scarce regions who might want a quick remote-sensing starting point. A reader already familiar with object-based image analysis would not learn new methods but could see one more city-level example. I would send it for peer review so the full methods and any validation results can be examined; the application has enough potential utility that referees should check whether the missing quantitative support is actually there in the manuscript.

Referee Report

3 major / 0 minor

Summary. The paper develops an automatic classification pipeline for road pavement types (paved vs. unpaved) in Greater Maputo using satellite images. It employs an object-oriented approach that classifies each road segment according to the distribution of its pixels in RGB space and asserts that the resulting method is accurate, inexpensive, and readily replicable in other cities.

Significance. If the accuracy claim holds after proper validation, the work would offer a low-cost, scalable approach to filling critical gaps in road network databases for developing countries, supporting better mobility planning with publicly available satellite data.

major comments (3)

[Abstract] Abstract: the assertion that the proposed approach is 'proven to be accurate' is unsupported by any quantitative results, validation details, error rates, confusion matrices, or baseline comparisons, rendering the central claim unevaluable from the manuscript.
[Abstract] Abstract: the method is described solely as 'looking at the distribution of its pixels in the RGB space' with no specification of the classification procedure (statistical measures, thresholds, models, or decision rules), preventing assessment of reproducibility or robustness.
[Abstract] Abstract: the approach does not address or test for common satellite-image confounders (illumination variation, shadows, vegetation, atmospheric haze, or resolution limits) that can cause overlapping RGB distributions between paved and unpaved surfaces, undermining the weakest assumption that RGB histograms alone suffice for reliable separation.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive comments. We address each major comment below and revise the abstract to improve clarity, support claims with evidence from the full manuscript, and acknowledge limitations.

read point-by-point responses

Referee: [Abstract] Abstract: the assertion that the proposed approach is 'proven to be accurate' is unsupported by any quantitative results, validation details, error rates, confusion matrices, or baseline comparisons, rendering the central claim unevaluable from the manuscript.

Authors: We agree that the abstract claim requires supporting details to be evaluable. The full manuscript contains the validation results from Greater Maputo; we will revise the abstract to summarize key quantitative metrics such as accuracy rates and any baseline comparisons. revision: yes
Referee: [Abstract] Abstract: the method is described solely as 'looking at the distribution of its pixels in the RGB space' with no specification of the classification procedure (statistical measures, thresholds, models, or decision rules), preventing assessment of reproducibility or robustness.

Authors: We acknowledge the abstract is too vague on the procedure. We will revise it to briefly specify the classification approach, including the statistical measures or decision rules applied to the RGB pixel distributions for each road segment. revision: yes
Referee: [Abstract] Abstract: the approach does not address or test for common satellite-image confounders (illumination variation, shadows, vegetation, atmospheric haze, or resolution limits) that can cause overlapping RGB distributions between paved and unpaved surfaces, undermining the weakest assumption that RGB histograms alone suffice for reliable separation.

Authors: This highlights a valid limitation of the abstract. The full manuscript applies the method to real-world satellite imagery in Greater Maputo, which includes such conditions, but we will revise the abstract to explicitly note the assumption and add a sentence on robustness while expanding the limitations discussion in the main text. revision: yes

Circularity Check

0 steps flagged

No circularity: empirical RGB histogram classification with no derivations or fitted predictions

full rationale

The paper describes an object-oriented pipeline that classifies road segments directly from the empirical distribution of RGB pixel values in satellite imagery. No equations, parameter estimation steps, uniqueness theorems, or self-citations appear in the provided abstract or reader summary. The central claim is a direct empirical procedure rather than any reduction of a 'prediction' to its own fitted inputs. This matches the default expectation of a non-circular applied-methods paper.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The approach rests on the untested domain assumption that RGB distributions alone separate pavement classes. No free parameters, invented entities, or additional axioms are stated in the abstract.

axioms (1)

domain assumption RGB pixel distributions extracted from satellite imagery are sufficient to discriminate paved from unpaved road surfaces.
Central to the object-oriented classification step described in the abstract.

pith-pipeline@v0.9.0 · 5609 in / 1125 out tokens · 48228 ms · 2026-05-23T20:59:03.349883+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

15 extracted references · 15 canonical work pages

[1]

doi: 10.3390/rs12091444

ISSN 2072–4292. doi: 10.3390/rs12091444. R. Behrens, D. McCormick, and D. Mfinanga. Paratransit in African Cities: Opera- tions, Regulation and Reform . Routledge,

work page doi:10.3390/rs12091444 2072
[2]

doi: 10.1109/TIT.1970.1054406. M. Ester, H.-P. Kriegel, J. Sander, and X. Xu. A Density-Based Algorithm for Dis- covering Clusters in Large Spatial Databases with Noise. In KDD, volume 96, pages 226–231. Association for the Advancement of Artificial Intelligence AIII Press,

work page doi:10.1109/tit.1970.1054406 1970
[3]

Accessed: 2020/04/20

URL http://www.google.com/earth/index.html. Accessed: 2020/04/20. C. Hennig. fpc: Flexible Procedures for Clustering ,

work page 2020
[4]

IEEE Transactions on Image Processing 26(5), 2274–2285 (2017)

doi: 10.1109/TIP. 2016.2559803. 18 D. P. Huttenlocher, G. A. Klanderman, and W. J. Rucklidge. Comparing images using the Hausdorff distance. IEEE Transactions on Pattern Analysis and Machine Intelligence, 15(9):850–863,

work page doi:10.1109/tip 2016
[5]

doi: 10.1109/34.232073. JICA. O Estudo Preparat´ orio sobre Plano de Melhoramenro da Estrada no Corredor de Desenvolvimento de Nacala (N13: Cuamba-Mandimba-Lichinga) na Rep´ ublica de Mo¸ cambique. Final report, Administra¸ c˜ ao Nacional de Estradas - Rep´ ublica de Mo¸ cambique,

work page doi:10.1109/34.232073
[6]

doi: 10.1109/ICAICA.2019.8873458. S. Marianingsih, F. Utaminingrum, and F. A. Bachtiar. Road Surface types classifi- cation using combination of K-nearest neighbor and Na¨ ıve Bayes based on GLCM. International Journal of Advances in Soft Computing and its Applications , 11(2): 15–27,

work page doi:10.1109/icaica.2019.8873458 2019
[7]

doi: https://doi.org/10.1002/bimj.201300072. J. S. Marron and I. L. Dryden. Object Oriented Data Analysis . Chapman and Hall/CRC, 1st edition,

work page doi:10.1002/bimj.201300072
[8]

Accessed: 2021/08/21

URL https://wiki.openstreetmap.org. Accessed: 2021/08/21. V. M. Panaretos and Y. Zemel. Statistical aspects of Wasserstein distances. An- nual review of statistics and its application , 6:405–431,

work page 2021
[9]

doi: 10.1121/1.3466870. I. Pillai, G. Fumera, and F. Roli. Multi-label classification with a reject option.Pattern Recognition, 46(8):2256–2266,

work page doi:10.1121/1.3466870
[10]

doi: https://doi.org/10.1016/j.patcog.2013.01

work page doi:10.1016/j.patcog.2013.01 2013
[11]

doi: 10.3390/infrastructures3040058. A. Riid, D. L. Manna, and S. Astapov. Image-based pavement type classification with convolutional neural networks. In 2020 IEEE 24th International Conference on Intelligent Engineering Systems , pages 55–60,

work page doi:10.3390/infrastructures3040058 2020
[12]

2020.9147199

doi: 10.1109/INES49302. 2020.9147199. M. Rizzo and G. Szekely. energy: E-Statistics: Multivariate Inference via the Energy of Data ,

work page doi:10.1109/ines49302 2020
[13]

Safari Njema,

URL http://www.jstor.org/stable/24306466. Safari Njema,

work page arXiv
[14]

Accessed: 2021/08/30

URL https://www.safari-njema.polimi.it/. Accessed: 2021/08/30. D. Schuhmacher, B. B¨ ahre, C. Gottschlich, V. Hartmann, F. Heinemann, and B. Schmitzer. transport: Computation of Optimal Transport Plans and Wasser- stein Distances,

work page 2021
[15]

doi: 10.1109/ICPR.2014.409. G. J. Sz´ ekely and M. L. Rizzo. Testing for equal distributions in high dimension. InterStat, 5(16.10):1249–1272,

work page doi:10.1109/icpr.2014.409 2014

[1] [1]

doi: 10.3390/rs12091444

ISSN 2072–4292. doi: 10.3390/rs12091444. R. Behrens, D. McCormick, and D. Mfinanga. Paratransit in African Cities: Opera- tions, Regulation and Reform . Routledge,

work page doi:10.3390/rs12091444 2072

[2] [2]

doi: 10.1109/TIT.1970.1054406. M. Ester, H.-P. Kriegel, J. Sander, and X. Xu. A Density-Based Algorithm for Dis- covering Clusters in Large Spatial Databases with Noise. In KDD, volume 96, pages 226–231. Association for the Advancement of Artificial Intelligence AIII Press,

work page doi:10.1109/tit.1970.1054406 1970

[3] [3]

Accessed: 2020/04/20

URL http://www.google.com/earth/index.html. Accessed: 2020/04/20. C. Hennig. fpc: Flexible Procedures for Clustering ,

work page 2020

[4] [4]

IEEE Transactions on Image Processing 26(5), 2274–2285 (2017)

doi: 10.1109/TIP. 2016.2559803. 18 D. P. Huttenlocher, G. A. Klanderman, and W. J. Rucklidge. Comparing images using the Hausdorff distance. IEEE Transactions on Pattern Analysis and Machine Intelligence, 15(9):850–863,

work page doi:10.1109/tip 2016

[5] [5]

doi: 10.1109/34.232073. JICA. O Estudo Preparat´ orio sobre Plano de Melhoramenro da Estrada no Corredor de Desenvolvimento de Nacala (N13: Cuamba-Mandimba-Lichinga) na Rep´ ublica de Mo¸ cambique. Final report, Administra¸ c˜ ao Nacional de Estradas - Rep´ ublica de Mo¸ cambique,

work page doi:10.1109/34.232073

[6] [6]

doi: 10.1109/ICAICA.2019.8873458. S. Marianingsih, F. Utaminingrum, and F. A. Bachtiar. Road Surface types classifi- cation using combination of K-nearest neighbor and Na¨ ıve Bayes based on GLCM. International Journal of Advances in Soft Computing and its Applications , 11(2): 15–27,

work page doi:10.1109/icaica.2019.8873458 2019

[7] [7]

doi: https://doi.org/10.1002/bimj.201300072. J. S. Marron and I. L. Dryden. Object Oriented Data Analysis . Chapman and Hall/CRC, 1st edition,

work page doi:10.1002/bimj.201300072

[8] [8]

Accessed: 2021/08/21

URL https://wiki.openstreetmap.org. Accessed: 2021/08/21. V. M. Panaretos and Y. Zemel. Statistical aspects of Wasserstein distances. An- nual review of statistics and its application , 6:405–431,

work page 2021

[9] [9]

doi: 10.1121/1.3466870. I. Pillai, G. Fumera, and F. Roli. Multi-label classification with a reject option.Pattern Recognition, 46(8):2256–2266,

work page doi:10.1121/1.3466870

[10] [10]

doi: https://doi.org/10.1016/j.patcog.2013.01

work page doi:10.1016/j.patcog.2013.01 2013

[11] [11]

doi: 10.3390/infrastructures3040058. A. Riid, D. L. Manna, and S. Astapov. Image-based pavement type classification with convolutional neural networks. In 2020 IEEE 24th International Conference on Intelligent Engineering Systems , pages 55–60,

work page doi:10.3390/infrastructures3040058 2020

[12] [12]

2020.9147199

doi: 10.1109/INES49302. 2020.9147199. M. Rizzo and G. Szekely. energy: E-Statistics: Multivariate Inference via the Energy of Data ,

work page doi:10.1109/ines49302 2020

[13] [13]

Safari Njema,

URL http://www.jstor.org/stable/24306466. Safari Njema,

work page arXiv

[14] [14]

Accessed: 2021/08/30

URL https://www.safari-njema.polimi.it/. Accessed: 2021/08/30. D. Schuhmacher, B. B¨ ahre, C. Gottschlich, V. Hartmann, F. Heinemann, and B. Schmitzer. transport: Computation of Optimal Transport Plans and Wasser- stein Distances,

work page 2021

[15] [15]

doi: 10.1109/ICPR.2014.409. G. J. Sz´ ekely and M. L. Rizzo. Testing for equal distributions in high dimension. InterStat, 5(16.10):1249–1272,

work page doi:10.1109/icpr.2014.409 2014