Pith. sign in

REVIEW 3 major objections 4 minor 22 references

SAMSelect: A Spectral Index Search for Marine Debris Visualization using Segment Anything

T0 review · 3 major / 4 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read SAMSelect searches 1,646 spectral band and index combinations to find the visualization that lets a segmentation model best match expert annotations, and finds a blue–NIR normalized difference index that outperforms NDVI and FDI for…

desk verdict A useful tool with an honest premise, but the headline IoU gains are in-sample maxima over the same annotated patches—held-out evaluation is needed before the generalizability claim holds. read the letter →

arxiv 2506.08613 v1 pith:7Z2OLTU4 submitted 2025-06-10 cs.CV

Joost van Dalen , Yuki M. Asano , Marc Russwurm This is my paper
classification cs.CV
keywords SAMSelectmarinedebrisspectralindicesSentinel-2SegmentAnythingModelbandselectionphoto-interpretationsegmentation
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

SAMSelect is an algorithm that picks the best three-channel false-color visualization of a multispectral satellite image by asking a large vision model to segment annotated objects and scoring each candidate visualization by how well the segmentation matches the annotation. Applied to Sentinel-2 images of marine debris in Accra and Durban, it searches 1,646 band and index combinations and finds that a normalized difference index using the blue (B2) and near-infrared (B8) bands makes debris far more visible than the standard NDVI or Floating Debris Index, raising IoU scores from the 9.6–27.7 range to 36.3–45.8. The paper's central assumption is that SAM's segmentation accuracy is a proxy for human visual interpretability, which would matter because marine scientists routinely interpret debris by eye and currently choose indices by heuristics.

What carries the argument

The central mechanism is the visualization function $V$ that maps a $D$-channel reflectance image to an RGB image through percentile normalization plus either a band composite, a normalized difference index $\mathrm{NDI}(b_1,b_2)=\frac{b_1-b_2}{b_1+b_2}$, a spectral shape index with linear interpolation between two flanking bands, or a composite of such indices (SIC). SAMSelect solves $V^\star = \arg\max_V \mathrm{IoU}\big(\mathrm{SAM}(V(X),p),\,Y\big)$ by exhaustive search over 1,646 candidates, using SAM with the ViT-B encoder as a proxy for human visual interpretation and deriving point prompts from expert annotations either manually or automatically.

What would settle it

A blind perceptual study in which expert photo-interpreters rank the SAMSelect-chosen visualizations (NDI(B2,B8) and the SIC) against NDVI, FDI, and true color; if expert preference does not track SAM's IoU ordering, the proxy assumption is false.

Watch

Extended reading notes

Core claim

The paper claims that optimal visualizations for a given annotation task can be discovered automatically rather than hand-picked. SAMSelect formulates this as maximizing $\mathrm{IoU}\big(\mathrm{SAM}(V(X),p),\,Y\big)$ over a visualization function $V$ chosen from four families (band composites, normalized difference indices, spectral shape indices, and spectral index composites), and implements the maximization by exhaustive search over 1,646 candidates on Sentinel-2 L2A data. On the Accra and Durban debris scenes, the selected indices—notably $\mathrm{NDI}(B2,B8)=\frac{B2-B8}{B2+B8}$ and SIC composites of NDI and SSI channels—consistently outperform NDVI, FDI, and a PCA baseline, with best IoU values of 45.8% (Accra) and 42.0% (Durban). The paper interprets this as showing that band combinations previously unused in marine-debris practice, especially the blue–NIR normalized difference, contain useful spectral contrast that domain heuristics had missed.

Load-bearing premise

The load-bearing premise is that the three-channel visualization that yields the most accurate SAM segmentations also gives the best visual information for human photo-interpretation; the paper never tests this against human experts.

Editorial extensions

If this is right

  • Marine scientists can compute $\mathrm{NDI}(B2,B8)$ directly as a single-channel visualization; in the tested scenes it outperformed NDVI and FDI.
  • Spectral Index Composites, which combine top NDI and SSI channels, produced the highest IoU (45.8% Accra, 42.0% Durban) by packing four or five bands into three RGB channels.
  • SAMSelect generalizes beyond marine debris and Sentinel-2 to other multispectral sensors and higher-dimensional inputs, though exhaustive search becomes expensive and would need Bayesian optimization.
  • The selected visualizations are robust to atmospheric correction: Sen2Cor and ACOLITE products differ by less than 1% IoU.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The paper evaluates selection and performance with the same IoU metric, so the claimed gains may not transfer to human photo-interpretation; a human perception study would test this.
  • The recurring choice of blue band B2 resonates with the Rotation-Absorption Index for oil slicks, hinting at a general blue/NIR contrast for floating material that a physically motivated index could exploit.
  • SAMSelect could be turned into a cheap screening tool for coastal plastic monitoring, but its band choices should be validated on more scenes and against in-situ targets before operational use.
  • Applying SAMSelect to hyperspectral or time-series data with Bayesian optimization could reveal narrow spectral or temporal signatures of different debris types.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 4 minor

Summary. The paper proposes SAMSelect, a method that automatically searches over three-channel visualizations of multispectral imagery (band composites, normalized difference indices, spectral shape indices, and their composites) by maximizing the IoU between SAM segmentations and user-provided polygon annotations. It is evaluated on two Sentinel-2 scenes with floating debris (Accra and Durban) plus a qualitative test on the Plastic Litter Project 2021 plastic targets. The authors report that the selected visualizations—notably NDI(B2,B8) and scene-specific spectral index composites—outperform established indices NDVI and FDI by 8.6% to 23% IoU, and they provide open-source code.

Significance. If the central claim were supported, SAMSelect would be a practically useful tool for marine scientists who visually interpret multispectral imagery, because it automates the currently manual, heuristic selection of band/index combinations. The paper has clear strengths: the search space is exhaustively enumerated, the code is open-source, runtime is reported, and the appendix contains ablations on point-prompt selection, SAM encoders, and atmospheric correction products. The qualitative PLP test is a sensible attempt at an external check. However, the quantitative evidence for the headline improvement is currently an in-sample selection result, and the core proxy assumption—that SAM IoU tracks human visual interpretability—is not independently validated. These issues are fixable within the scope of a revision, so the work is promising but not yet conclusive.

major comments (3)
  1. [Section IV-B, Table I, Eq. (1)] The quantitative comparison in Table I is subject to selection-evaluation confounding. SAMSelect's objective in Eq. (1) selects the visualization V* that maximizes IoU over the same annotated patches (5 for Accra, 16 for Durban) that are then used to report the IoU values in Table I after searching 1,646 candidate visualizations. The reported improvements of 8.6% to 23% over NDVI and FDI are therefore maxima of an exhaustive search on a small dataset, not unbiased estimates of performance on unseen data. This does not establish that NDI(B2,B8) or the SIC composites transfer to new scenes or even to unannotated patches within the same scene. The manuscript should either add a held-out evaluation (for example, a train/test split of patches, or cross-scene transfer of the Accra-found visualization to Durban and PLP with IoU reported) or explicitly reframe Table I as a per-scene calibration result rather than comparative evidence of improved performance.
  2. [Abstract, Section I, Section IV-A] The central assumption stated in the abstract and introduction—that the visualization achieving the most accurate SAM segmentation also provides the best visual information for photo-interpretation—is not validated. Table I only reports IoU, which is the same metric used for selection, so it cannot validate the proxy. The only external check, the PLP targets in Section IV-A, is qualitative and does not report IoU or any human interpretability measurement. The conclusion that SAMSelect yields 'salient' visualizations for marine scientists therefore rests on an untested premise. I recommend adding a human-subject evaluation (even a small expert survey) or a quantitative interpretability proxy (e.g., a contrast-to-noise metric) on the PLP targets.
  3. [Section IV and Section V] The claim that NDI(B2,B8) 'consistently produced most salient visualizations across both Accra, Durban and the PLP2021 targets' is based on only two annotated scenes plus a qualitative inspection. The full SIC composites differ between Accra and Durban, which indicates that the method is effectively per-scene calibration rather than discovery of a universal index. The generalization claim in the conclusion ('SAMSelect has consistently selected the NDI B2,B8 combination') needs quantitative cross-scene evidence: for example, report the IoU of the Accra-selected SIC on the Durban patches and vice versa. Without such numbers, the conclusion overstates the transferability of the findings.
minor comments (4)
  1. [Section II, SIC definition] In the definition of Spectral Index Composites, 'c green ← SI 2, c green ← SI 3' appears to be a typo; the third assignment should presumably be 'c blue ← SI 3'. Similarly, in the band composite example, 'c green ← b B2' should likely be 'c blue ← b B2'.
  2. [Section II, Search Space] The SIC search space is described as using the 'top-10 most informative NDI and SSIs', but the criterion for 'most informative' is not defined. For reproducibility, specify the ranking metric (e.g., variance, discriminant power, or mutual information with annotations).
  3. [Section II, Point Prompt Selector] The K-means prompt selection uses K=10, and the percentile scaling uses 1%-99% thresholds; both are said to be determined empirically. Please state explicitly whether these hyperparameters were tuned on the same annotated patches used for the final evaluation, as this would add another (minor) layer of selection bias.
  4. [Section IV-A and Table II] The qualitative statement that NDI(B2,B8) 'provides more pronounced contrast' could be quantified, for example with a simple mean-contrast or signal-to-background ratio on the PLP targets. Also, in Table II the column header 'Indices' lists 'BC' and 'NDI', so 'Indices' should be 'Visualization type'.

Circularity Check

1 steps flagged · score 6.0 of 10

Table I's IoU gains are in-sample maxima: SAMSelect selects V* by maximizing IoU on the same annotated patches used for evaluation, and the baselines NDVI/FDI are members of the searched space.

  1. fitted input called prediction [Section II, Eq. (1); Section IV-B, Table I and surrounding text]
    "V ⋆ = ... argV max [IoU(SAM(V(X),p), Y)] ... over a small dataset D ... We implement the maximization by testing an exhaustive search space of possible visualizations. ... These visualizations are compared against the literature-based indices of NDVI, FDI [7] ... Overall, the results demonstrate a quantitative improvement over established indices NDVI and FDI, with segmentation scores increasing within a range of 8.6% to 23%."

    Equation (1) defines the selected visualization V* as the argmax of IoU over the annotated dataset D. Table I reports the IoU of that same V* on the same D and presents it as a quantitative improvement over NDVI and FDI. Since NDVI is exactly NDI(B4,B8) and FDI is an SSI(B4,B8,B11), both lie inside the 1,646-candidate search space, so the searched maximum is guaranteed to score no worse than these baselines on the training patches. The reported 8.6-23% gains are therefore in-sample selection maxima, not unbiased estimates of how the chosen visualization would perform on unseen data or unannotated patches. The only out-of-sample check, the PLP 2021 targets, is used qualitatively and does not report IoU, so the quantitative claim reduces to the optimization objective by construction.

full rationale

The paper's central quantitative claim is statistically forced by its own selection rule. SAMSelect exhaustively searches visualizations and picks the one maximizing Eq. (1)'s IoU on the user-supplied annotations (5 patches for Accra, 16 for Durban). Table I then evaluates that same selected visualization on the same annotated patches and reports higher IoU than NDVI and FDI. Because both baseline indices are special cases of the searched NDI and SSI families, the selected best index or composite must meet or exceed them on the fitting data; the 8.6-23% margin is an artifact of the argmax over a large search space, not evidence of transfer. The paper is transparent that its 'central assumption' is that SAM accuracy proxies human interpretability, and the PLP example gives qualitative plausibility, but no quantitative IoU is reported for PLP, so the external check does not break the selection-evaluation loop. I do not find a load-bearing self-citation chain: the FloatingObjects dataset [11] is an external annotation source and FDI [7] is a literature baseline. The circularity is thus partial and located in the evaluation design rather than in the definition of the indices. Score 6: the reported improvements reduce by construction to the optimization objective, while the proposed tool itself retains some practical content as a per-scene calibration procedure.

Assumptions & free parameters 4 free parameters · 5 assumptions · 0 invented entities

The ledger records no invented physical entities. All main empirical results rest on a small two-scene annotated dataset, an unvalidated proxy assumption (SAM IoU equals interpretability), and several empirically chosen hyperparameters (percentile scaling, K=10 for K-means prompts, top-10 index preselection). The winning visualization itself is a fitted quantity.

free parameters (4)
  • Percentile scaling thresholds = 1st and 99th percentiles
    Chosen experimentally over histogram equalization and min-max scaling in Section II (Method).
  • K for K-means point prompt selection = 10
    Determined empirically in Section II (Point Prompt Selector): lower K misses objects, higher K distracts SAM.
  • Top-10 informative NDI/SSI preselect for SIC = 10
    SIC search uses only top-10 indices ranked by informativeness, a data-dependent filter imposed before the SIC evaluation.
  • Selected visualization V* = SIC Accra: NDI(B2,B8), SSI(B1,B8,B11), SSI(B2,B8,B11); SIC Durban: NDI(B2,B8), NDI(B1,B8A), NDI(B3,B8)
    The output itself is the argmax of IoU on the annotated evaluation scenes, and the quantitative gains are measured on this same dataset.
assumptions (5)
  • domain assumption SAM segmentation accuracy is a faithful proxy for human photo-interpretation quality.
    Central assumption stated in the abstract and Section I; not validated with human subjects.
  • domain assumption IoU between SAM prediction and provided annotations measures the usefulness of a visualization.
    Optimization objective Eq. (1) equates the best visualization with the highest IoU.
  • domain assumption FloatingObjects polygon annotations are accurate for debris locations.
    Used to supervise the search; no error analysis is provided for the annotations.
  • domain assumption Bilinear interpolation of 20m bands to 10m preserves spectral information relevant to debris.
    Data section says all bands were interpolated to 10m during GEE export; no analysis of interpolation artifacts is given.
  • domain assumption Exhaustive search over 1,646 candidate visualizations is sufficient to find globally salient combinations.
    Search space is limited to specific BC, NDI, SSI, and SIC families; other indices or channel assignments are excluded.

how reviews work

0 comments
Cite this review

Pith. "Pith review of SAMSelect: A Spectral Index Search for Marine Debris Visualization using Segment Anything." pith.science (2026). https://pith.science/paper/7Z2OLTU4

@misc{pith2026250608613,
  author       = {Pith},
  title        = {Pith review of: SAMSelect: A Spectral Index Search for Marine Debris Visualization using Segment Anything},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/7Z2OLTU4}},
  note         = {Machine review of arXiv:2506.08613}
}
read the original abstract

This work proposes SAMSelect, an algorithm to obtain a salient three-channel visualization for multispectral images. We develop SAMSelect and show its use for marine scientists visually interpreting floating marine debris in Sentinel-2 imagery. These debris are notoriously difficult to visualize due to their compositional heterogeneity in medium-resolution imagery. Out of these difficulties, a visual interpretation of imagery showing marine debris remains a common practice by domain experts, who select bands and spectral indices on a case-by-case basis informed by common practices and heuristics. SAMSelect selects the band or index combination that achieves the best classification accuracy on a small annotated dataset through the Segment Anything Model. Its central assumption is that the three-channel visualization achieves the most accurate segmentation results also provide good visual information for photo-interpretation. We evaluate SAMSelect in three Sentinel-2 scenes containing generic marine debris in Accra, Ghana, and Durban, South Africa, and deployed plastic targets from the Plastic Litter Project. This reveals the potential of new previously unused band combinations (e.g., a normalized difference index of B8, B2), which demonstrate improved performance compared to literature-based indices. We describe the algorithm in this paper and provide an open-source code repository that will be helpful for domain scientists doing visual photo interpretation, especially in the marine field.

Figures

Figures reproduced from arXiv: 2506.08613 by the authors.

Figure 1
Figure 1. Schematic of the SAMSelect algorithm, automating spectral band selection by maximizing the Intersection over Union [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. This result is notable given that the blue band is not a [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 2
Figure 2. Visualizations of Sentinel-2 images from Accra (2018- [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

22 extracted references · 19 canonical work pages

  1. [1]

    A unified vegetation index for quantifying the terrestrial biosphere,

    G. Camps-Valls, M. Campos-Taberner, ´A. Moreno-Mart´ınez, S. Walther, G. Duveiller, A. Cescatti, M. D. Mahecha, J. Mu ˜noz-Mar´ı, F. J. Garc´ıa- Haro, L. Guanteret al., “A unified vegetation index for quantifying the terrestrial biosphere,”Science Advances, vol. 7, no. 9, p. eabc7447, 2021

    work page 2021

  2. [2]

    Overview of the radiometric and biophysical performance of the modis vegetation indices,

    A. Huete, K. Didan, T. Miura, E. P. Rodriguez, X. Gao, and L. G. Ferreira, “Overview of the radiometric and biophysical performance of the modis vegetation indices,”Remote sensing of environment, vol. 83, no. 1-2, pp. 195–213, 2002

    work page 2002

  3. [3]

    A novel ocean color index to detect floating algae in the global oceans,

    C. Hu, “A novel ocean color index to detect floating algae in the global oceans,”Remote Sensing of Environment, vol. 113, no. 10, pp. 2118– 2129, 2009

    work page 2009

  4. [4]

    Normalized difference chlorophyll index: A novel model for remote estimation of chlorophyll-a concentration in turbid productive waters,

    S. Mishra and D. R. Mishra, “Normalized difference chlorophyll index: A novel model for remote estimation of chlorophyll-a concentration in turbid productive waters,”Remote Sensing of Environment, vol. 117, pp. 394–406, 2012

    work page 2012

  5. [5]

    Modified modis fluorescence line height data product to improve image interpretation for red tide monitoring in the eastern gulf of mexico,

    C. Hu and L. Feng, “Modified modis fluorescence line height data product to improve image interpretation for red tide monitoring in the eastern gulf of mexico,”Journal of Applied Remote Sensing, vol. 11, no. 1, pp. 012 003–012 003, 2017

    work page 2017

  6. [6]

    Investigating detection of floating plastic litter from space using sentinel-2 imagery,

    K. Themistocleous, C. Papoutsa, S. Michaelides, and D. Hadjimitsis, “Investigating detection of floating plastic litter from space using sentinel-2 imagery,”Remote Sensing, vol. 12, no. 16, p. 2648, 2020

    work page 2020

  7. [7]

    Finding plastic patches in coastal waters using optical satellite data,

    L. Biermann, D. Clewley, V . Martinez-Vicente, and K. Topouzelis, “Finding plastic patches in coastal waters using optical satellite data,” Scientific reports, vol. 10, no. 1, p. 5364, 2020

    work page 2020

  8. [8]

    Segment anything,

    A. Kirillov, E. Mintun, N. Ravi, H. Mao, C. Rolland, L. Gustafson, T. Xiao, S. Whitehead, A. C. Berg, W.-Y . Loet al., “Segment anything,” inProceedings of the IEEE/CVF International Conference on Computer Vision, 2023, pp. 4015–4026

    2023

Show all 22 references
  1. [9]

    Ndwi—a normalized difference water index for remote sensing of vegetation liquid water from space,

    B.-C. Gao, “Ndwi—a normalized difference water index for remote sensing of vegetation liquid water from space,”Remote sensing of environment, vol. 58, no. 3, pp. 257–266, 1996

    1996

  2. [10]

    The use of the normalized difference water index (ndwi) in the delineation of open water features,

    S. K. McFeeters, “The use of the normalized difference water index (ndwi) in the delineation of open water features,”International journal of remote sensing, vol. 17, no. 7, pp. 1425–1432, 1996

    1996

  3. [11]

    Large-scale detection of marine debris in coastal areas with sentinel-2,

    M. Rußwurm, S. J. Venkatesa, and D. Tuia, “Large-scale detection of marine debris in coastal areas with sentinel-2,” vol. 26, no. 12, p. 108402. [Online]. Available: https://linkinghub.elsevier.com/retrieve/pii/ S2589004223024793

  4. [12]

    A fast parallel algorithm for thinning digital patterns,

    T. Y . Zhang and C. Y . Suen, “A fast parallel algorithm for thinning digital patterns,”Communications of the ACM, vol. 27, no. 3, pp. 236–239, Mar. 1984. [Online]. Available: https://dl.acm.org/doi/10. 1145/357994.358023

    1984

  5. [13]

    An Empirical Assessment of Marine Debris, Seawater Quality and Littering in Ghana,

    I. P. V . Dyck, F. K. E. Nunoo, and E. T. Lawson, “An Empirical Assessment of Marine Debris, Seawater Quality and Littering in Ghana,”Journal of Geoscience and Environment Protection, vol. 4, no. 5, pp. 21–36, May 2016, number: 5 Publisher: Scientific Research Publishing. [Onl...

    2016

  6. [14]

    Exploring plastic transport dynamics in the Odaw river, Ghana,

    R. B. Pinto, T. Barendse, T. van Emmerik, M. van der Ploeg, F. O. Annor, K. Duah, J. Udo, and R. Uijlenhoet, “Exploring plastic transport dynamics in the Odaw river, Ghana,”Frontiers in Environmental Science, vol. 11, Feb. 2023, publisher: Frontiers. [Online]. Available: https...

    2023

  7. [15]

    The impact of floods on plastic pollution,

    T. H. Van Emmerik, “The impact of floods on plastic pollution,”Global Sustainability, vol. 7, p. e17, 2024, publisher: Cambridge University Press. [Online]. Available: https://www.cambridge.org/core/journals/ global-sustainability/article/impact-of-floods-on-plastic-pollution/...

    2024

  8. [16]

    Proof of concept for a new sensor to monitor marine litter from space,

    A. C ´ozar, M. Arias, G. Suaria, J. Viejo, S. Aliani, A. Koutroulis, J. Delaney, G. Bonnery, D. Mac ´ıas, R. de Vries, R. Sumerot, C. Morales-Caselles, A. Turiel, D. Gonz ´alez-Fern´andez, and P. Corradi, “Proof of concept for a new sensor to monitor marine litter from space,”...

    2024

  9. [17]

    Sentinel-2 Detection of Floating Marine Litter Targets with Partial Spectral Unmixing and Spectral Comparison with Other Floating Materials (Plastic Litter Project 2021),

    D. Papageorgiou, K. Topouzelis, G. Suaria, S. Aliani, and P. Corradi, “Sentinel-2 Detection of Floating Marine Litter Targets with Partial Spectral Unmixing and Spectral Comparison with Other Floating Materials (Plastic Litter Project 2021),”Remote Sensing, vol. 14, no. 23, p....

    2021

  10. [18]

    Charac- terization of oil slicks at sea using remote sensing techniques,

    E. Loos, L. Brown, G. Borstad, T. Mudge, and M. ´Alvarez, “Charac- terization of oil slicks at sea using remote sensing techniques,” in2012 Oceans. IEEE, 2012, pp. 1–4

    2012

  11. [19]

    Bayesian Optimization: Open source constrained global optimization tool for Python,

    F. Nogueira, “Bayesian Optimization: Open source constrained global optimization tool for Python,” 2014–. [Online]. Available: https://github.com/bayesian-optimization/BayesianOptimization

    2014

  12. [20]

    Sen2Cor for Sentinel-2,

    M. Main-Knorn, B. Pflug, J. Louis, V . Debaecker, U. M ¨uller-Wilm, and F. Gascon, “Sen2Cor for Sentinel-2,” vol. 10427, p. 1042704, Oct. 2017, conference Name: Image and Signal Processing for Remote Sensing XXIII ADS Bibcode: 2017SPIE10427E..04M. [Online]. Available: https://...

    2017

  13. [21]

    Adaptation of the dark spectrum fitting atmospheric correction for aquatic applications of the Landsat and Sentinel-2 archives,

    Q. Vanhellemont, “Adaptation of the dark spectrum fitting atmospheric correction for aquatic applications of the Landsat and Sentinel-2 archives,”Remote Sensing of Environment, vol. 225, pp. 175–192, May 2019. [Online]. Available: https://www.sciencedirect.com/science/ article...

    2019

  14. [22]

    Detection of floating plastics from satellite and unmanned aerial systems (Plastic Litter Project 2018),

    K. Topouzelis, A. Papakonstantinou, and S. P. Garaba, “Detection of floating plastics from satellite and unmanned aerial systems (Plastic Litter Project 2018),”International Journal of Applied Earth Observation and Geoinformation, vol. 79, pp. 175–183, Jul. 2019. [Online]. Ava...

    2018

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.