Investigating the Segment Anything Foundation Model for Mapping Smallholder Agriculture Field Boundaries Without Training Labels

Jyles Watson; Kathy Baylis; Kyle Wu; Pratyush Tripathy; Ruizhe Jiang

arxiv: 2407.01846 · v1 · submitted 2024-07-01 · 💻 cs.CV

Investigating the Segment Anything Foundation Model for Mapping Smallholder Agriculture Field Boundaries Without Training Labels

Pratyush Tripathy , Kathy Baylis , Kyle Wu , Jyles Watson , Ruizhe Jiang This is my paper

Pith reviewed 2026-05-23 22:56 UTC · model grok-4.3

classification 💻 cs.CV

keywords segment anything modelfield boundary mappingsatellite imageryzero-shot segmentationsmallholder agricultureBihar Indiaagricultural mapping

0 comments

The pith

The Segment Anything Model identifies about 58% of agricultural field boundaries from satellite imagery without any training data.

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

This paper tests whether the Segment Anything Model can delineate smallholder farm boundaries in satellite images of Bihar, India, without additional training. It applies SAM to 2-meter SkySat imagery across different checkpoints, sizes, and multi-date inputs. The model achieves roughly 58 percent accuracy on field boundaries, matching supervised methods that need lots of labeled data. Improvements come from using multiple dates and edge-enhanced images. The work shows SAM's potential for mapping in places lacking training datasets.

Core claim

The Segment Anything Model, when applied zero-shot to high-resolution satellite imagery, correctly identifies about 58% of field boundaries in smallholder agricultural landscapes, with performance improving notably when multi-date images are used as input.

What carries the argument

The Segment Anything Model (SAM) used in zero-shot mode on satellite imagery for boundary delineation.

Load-bearing premise

The 58% accuracy figure depends on the quality and independence of an unspecified ground-truth dataset used for evaluation.

What would settle it

An independent, high-quality ground-truth map of field boundaries in the same Bihar area that shows accuracy well below or above 58% when using the same SAM configuration.

Figures

Figures reproduced from arXiv: 2407.01846 by Jyles Watson, Kathy Baylis, Kyle Wu, Pratyush Tripathy, Ruizhe Jiang.

**Figure 1.** Figure 1: SkySat 2 m resolution satellite images for Bihar, India for dates, (d) 11 May 2015, (e) 29 September 2015, (f) 16 November 2015 and (g) 10 Feb 2016. [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗

**Figure 2.** Figure 2: Schematic diagram of the different levels of accuracy assessment for predicted layers. 2.9. Investigating different sizes of input image Since SAM can be run on images of any size, we systematically investigate the accuracy associated with different input image sizes. We clip the images for all four time periods into smaller tiles of four sizes: 256 x 256, 512 x 512, 768 x 768, and 1024 x 1024 pixels. We … view at source ↗

**Figure 4.** Figure 4: Accuracy metrics at level 2 (different image dates after combining checkpoints) and level 3 (combining image sizes) [PITH_FULL_IMAGE:figures/full_fig_p005_4.png] view at source ↗

**Figure 5.** Figure 5: Accuracy metrics for original and edge enhanced images at di [PITH_FULL_IMAGE:figures/full_fig_p005_5.png] view at source ↗

**Figure 6.** Figure 6: Histogram and scatter plot of area of actual and predicted field bound [PITH_FULL_IMAGE:figures/full_fig_p006_6.png] view at source ↗

read the original abstract

Accurate mapping of agricultural field boundaries is crucial for enhancing outcomes like precision agriculture, crop monitoring, and yield estimation. However, extracting these boundaries from satellite images is challenging, especially for smallholder farms and data-scarce environments. This study explores the Segment Anything Model (SAM) to delineate agricultural field boundaries in Bihar, India, using 2-meter resolution SkySat imagery without additional training. We evaluate SAM's performance across three model checkpoints, various input sizes, multi-date satellite images, and edge-enhanced imagery. Our results show that SAM correctly identifies about 58% of field boundaries, comparable to other approaches requiring extensive training data. Using different input image sizes improves accuracy, with the most significant improvement observed when using multi-date satellite images. This work establishes proof of concept for using SAM and maximizing its potential in agricultural field boundary mapping. Our work highlights SAM's potential in delineating agriculture field boundary in training-data scarce settings to enable a wide range of agriculture related analysis.

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

SAM zero-shot on smallholder field boundaries reports 58% but the number is hard to interpret without a defined metric or independent ground truth.

read the letter

The paper applies the Segment Anything Model zero-shot to delineate field boundaries from 2m SkySat imagery in Bihar without any training labels. It tests three checkpoints, different input sizes, multi-date stacks, and edge-enhanced images, then reports that SAM correctly identifies about 58% of boundaries and that multi-date data helps most. This is the first reported use of SAM for this exact task in smallholder settings, so the application itself is new even if the underlying model is not. The experiments on input variations are straightforward and show some practical thinking about how to get more out of the model in an agricultural context. The multi-date improvement is the clearest takeaway and aligns with what one would expect from seasonal field changes. The central claim is an empirical report rather than a derivation, and the authors are explicit that this is a proof-of-concept for data-scarce regions. The main weakness is the 58% figure and the comparability statement. The abstract gives no information on the matching rule, the source or quality of the ground-truth boundaries, the number of fields or images evaluated, or the exact baselines used for comparison. Without those details it is impossible to judge whether the protocol is independent or whether the metric favors SAM's output style. If the full methods section supplies a clear, pre-specified evaluation with independent annotations, the result becomes more usable; if not, the headline number remains too underspecified to rely on. This is aimed at remote-sensing and agricultural analysts who need zero-shot options in places where labels are expensive. A reader looking for ideas on foundation-model applications would get some value from the input-variation tests, but anyone needing reproducible performance numbers should treat the 58% claim as preliminary. It deserves peer review if the methods section fills in the evaluation protocol; otherwise the main result cannot be assessed properly.

Referee Report

1 major / 2 minor

Summary. The paper investigates applying the Segment Anything Model (SAM) to delineate smallholder agricultural field boundaries from 2 m SkySat imagery in Bihar, India, without any task-specific training or fine-tuning. It reports results across three SAM checkpoints, varying input sizes, multi-date imagery, and edge-enhanced inputs, claiming that SAM correctly identifies about 58% of field boundaries and performs comparably to supervised methods that require extensive labeled data. The work positions this as a proof-of-concept for foundation models in data-scarce agricultural mapping.

Significance. If the 58% figure is supported by a clearly documented, independent ground-truth dataset and a reproducible evaluation protocol, the result would be significant for remote-sensing applications in smallholder systems: it would show that a general-purpose segmentation foundation model can deliver usable boundary maps without the labeling burden that currently limits supervised approaches in many regions.

major comments (1)

[Results] Results section (and abstract): The headline claim that SAM 'correctly identifies about 58% of field boundaries' is presented without any accompanying information on (a) the size and source of the ground-truth dataset, (b) the precise boundary-matching rule or tolerance used to declare a detection correct, or (c) quantitative baselines and error bars. This information is load-bearing for both the 58% number and the comparability statement; its absence prevents verification that the evaluation is independent of SAM outputs and that the metric is not inadvertently favorable to SAM's output style.

minor comments (2)

[Abstract] The abstract and introduction would benefit from a short statement of the total number of images and fields evaluated.
[Methods] Notation for the three SAM checkpoints (e.g., ViT-H, ViT-L, ViT-B) should be introduced consistently when first mentioned.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the constructive feedback on our manuscript. We agree that the evaluation details supporting the 58% figure require explicit documentation to allow independent verification, and we will revise the manuscript to address this.

read point-by-point responses

Referee: [Results] Results section (and abstract): The headline claim that SAM 'correctly identifies about 58% of field boundaries' is presented without any accompanying information on (a) the size and source of the ground-truth dataset, (b) the precise boundary-matching rule or tolerance used to declare a detection correct, or (c) quantitative baselines and error bars. This information is load-bearing for both the 58% number and the comparability statement; its absence prevents verification that the evaluation is independent of SAM outputs and that the metric is not inadvertently favorable to SAM's output style.

Authors: We acknowledge that the current version of the manuscript does not provide sufficient detail on these points in the Results section or abstract. In the revised manuscript we will add a clear description of (a) the ground-truth dataset (its size, collection method, and source, collected independently of any SAM outputs), (b) the exact boundary-matching protocol and tolerance used to score a detection as correct, and (c) quantitative comparisons to supervised baselines together with error bars or confidence intervals. These additions will be placed in the Methods and Results sections to ensure the 58% figure and comparability claims are fully reproducible and verifiable. revision: yes

Circularity Check

0 steps flagged

No circularity: empirical report with no derivations or fitted predictions

full rationale

The paper is an empirical evaluation of the Segment Anything Model on satellite imagery for field boundary detection, reporting a measured accuracy of ~58% without any equations, parameter fitting, or claimed first-principles derivations. The central claim is a direct performance observation on external imagery rather than a quantity constructed from quantities defined inside the paper. No self-citation chains, ansatzes, or uniqueness theorems are invoked to support a derivation, so the result does not reduce to its own inputs by construction.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests on the domain assumption that SAM's segmentation capability transfers to satellite imagery of smallholder fields without domain-specific fine-tuning or prompt engineering beyond the variations tested.

axioms (1)

domain assumption SAM generalizes zero-shot to satellite imagery of agricultural fields
Invoked when the authors apply the model without training and interpret the 58% result as evidence of utility.

pith-pipeline@v0.9.0 · 5707 in / 1104 out tokens · 21441 ms · 2026-05-23T22:56:30.996033+00:00 · methodology

discussion (0)

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

IndisputableMonolith/Cost/FunctionalEquation.lean washburn_uniqueness_aczel unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

We use the Intersection Over Union (IoU) metric... detection accuracy... proportion of predicted fields with an IoU value greater than 0.5... delineation accuracy... mean IoU value
IndisputableMonolith/Foundation/RealityFromDistinction.lean reality_from_one_distinction unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

Our results show that SAM correctly identifies about 58% of field boundaries

What do these tags mean?

matches: The paper's claim is directly supported by a theorem in the formal canon.
supports: The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends: The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses: The paper appears to rely on the theorem as machinery.
contradicts: The paper's claim conflicts with a theorem or certificate in the canon.
unclear: Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

17 extracted references · 17 canonical work pages

[1]

doi: https: //doi.org/10.1016/ j.rse.2016.03.010

ISSN 0034-4257. doi: https: //doi.org/10.1016/ j.rse.2016.03.010. URL https://www.sciencedirect.com/science/ article/pii/S0034425716301031. Lyndon D. Estes, Su Ye, Lei Song, Boka Luo, J. Ronald Eastman, Zhenhua Meng, Qi Zhang, Dennis McRitchie, Stephanie R. Debats, Justus Muhando, Angeline H. Amukoa, Brian W. Kaloo, Jackson Makuru, Ben K. Mbatia, Isaac M....

work page 2016
[2]

doi: 10.3389 /frai.2021

ISSN 2624-8212. doi: 10.3389 /frai.2021. 744863. URL https://www.frontiersin.org/articles/10.3389/ frai.2021.744863. Ashley Fernandez and Swaraj Dube. Core building blocks: Next gen geo spatial GPT application,

work page arXiv 2021
[3]

Artur Gafurov

URL http://arxiv.org/abs/2310.11029. Artur Gafurov. Automated Mapping of Cropland Boundaries Using Deep Neu- ral Networks. AgriEngineering, 5(3):1568–1580, September

work page arXiv
[4]

doi: 10.1016 /j.foodres.2014.03.015

ISSN 0963-9969. doi: 10.1016 /j.foodres.2014.03.015. URL https://www. sciencedirect.com/science/article/pii/S0963996914001732. Angel Garcia-Pedrero, Consuelo Gonzalo-Martin, Mario Lillo-Saavedra, and Dionisio Rodr´ıguez-Esparrag´on. The Outlining of Agricultural Plots Based on Spatiotemporal Consensus Segmentation. Remote Sensing, 10(12):1991, December

work page 2014
[5]

doi: 10.3390 /rs10121991

ISSN 2072-4292. doi: 10.3390 /rs10121991. URL http: //www.mdpi.com/2072-4292/10/12/1991. Johannes Jakubik, Sujit Roy, C. E. Phillips, Paolo Fraccaro, Denys Godwin, Bianca Zadrozny, Daniela Szwarcman, Carlos Gomes, Gabby Nyirjesy, Blair Edwards, Daiki Kimura, Naomi Simumba, Linsong Chu, S. Karthik Mukkavilli, Devyani Lambhate, Kamal Das, Ranjini Bangalore,...

work page 2072
[6]

Available: https://arxiv.org/abs/2310.18660

URL http://arxiv. org/abs/2310.18660. Alexander Kirillov, Eric Mintun, Nikhila Ravi, Hanzi Mao, Chloe Rolland, Laura Gustafson, Tete Xiao, Spencer Whitehead, Alexander C. Berg, Wan- Yen Lo, Piotr Doll´ar, and Ross Girshick. Segment anything,

work page arXiv
[7]

Monitoring forest dynamics in Africa during 2000 – 2020 using a remotely sensed fractional tree cover dataset,

doi: 10.1080/17538947.2023.2278895. URL https://doi.org/10.1080/ 17538947.2023.2278895. Weiye Mei, Haoyu Wang, David Fouhey, Weiqi Zhou, Isabella Hinks, Josh M. Gray, Derek Van Berkel, and Meha Jain. Using deep learning and very-high- resolution imagery to map smallholder field boundaries.Remote Sensing, 14 (13),

work page doi:10.1080/17538947.2023.2278895 2023
[8]

doi: 10.3390 /rs14133046

ISSN 2072-4292. doi: 10.3390 /rs14133046. Heather C North, David Pairman, and Stella E Belliss. Boundary delineation of agricultural fields in multitemporal satellite imagery. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing , 12(1): 237–251,

work page 2072
[9]

doi: 10.1038 /s41467-021-24638-z

ISSN 2041-1723. doi: 10.1038 /s41467-021-24638-z. URL https://www.nature.com/ articles/s41467-021-24638-z . Publisher: Nature Publishing Group. RB Singh, Praduman Kumar, T Woodhead, et al. Smallholder farmers in india: Food security and agricultural policy. Smallholder farmers in India: Food 7 security and agricultural policy,

work page 2041
[10]

doi: 10.3390/rs13040722

ISSN 2072-4292. doi: 10.3390/rs13040722. URL https://www.mdpi.com/2072-4292/13/4/

work page doi:10.3390/rs13040722 2072
[11]

doi: 10.3390 / rs12071205

ISSN 2072-4292. doi: 10.3390 / rs12071205. URL https://www.mdpi.com/2072-4292/12/7/1205. Franc ¸ois Waldner and Foivos I. Diakogiannis. Deep learning on edge: Ex- tracting field boundaries from satellite images with a convolutional neu- ral network. Remote Sensing of Environment , 245:111741, August

work page 2072
[12]

doi: 10.1016 /j.rse.2020.111741

ISSN 0034-4257. doi: 10.1016 /j.rse.2020.111741. URL https://www. sciencedirect.com/science/article/pii/S0034425720301115. Mo Wang, Jing Wang, Yunpeng Cui, Juan Liu, and Li Chen. Agricul- tural Field Boundary Delineation with Satellite Image Segmentation for High-Resolution Crop Mapping: A Case Study of Rice Paddy. Agron- omy, 12(10):2342, September 2022a...

work page doi:10.3390/rs14225738 2020
[13]

URL https://doi.org/10.1080/17538947.2024.2353122. M. Weiss, F. Jacob, and G. Duveiller. Remote sensing for agricultural applications: A meta-review. Remote Sensing of Environment , 236: 111402,

work page doi:10.1080/17538947.2024.2353122 2024
[14]

doi: https: //doi.org/10.1016/j.rse.2019

ISSN 0034-4257. doi: https: //doi.org/10.1016/j.rse.2019. 111402. URL https://www.sciencedirect.com/science/article/ pii/S0034425719304213. Qiusheng Wu and Lucas Prado Osco. samgeo: A python package for seg- menting geospatial data with the segment anything model (sam). Journal of Open Source Software, 8(89):5663,

work page doi:10.1016/j.rse.2019 2019
[15]

URL https://doi.org/10.21105/joss.05663

doi: 10.21105 /joss.05663. URL https://doi.org/10.21105/joss.05663. Ruoyu Yang, Zia U. Ahmed, Urs C. Schulthess, Mustafa Kamal, and Rahul Rai. Detecting functional field units from satellite images in smallholder farming systems using a deep learning based computer vision approach: A case study from bangladesh. Remote Sensing Applications: Society and Env...

work page doi:10.21105/joss.05663
[16]

doi: https: //doi.org/ 10.1016/j.rsase.2020.100413

ISSN 2352-9385. doi: https: //doi.org/ 10.1016/j.rsase.2020.100413. URL https://www.sciencedirect.com/ science/article/pii/S2352938519301831. Peng Zhang, Shougeng Hu, Weidong Li, and Chuanrong Zhang. Parcel-level mapping of crops in a smallholder agricultural area: A case of central china using single-temporal vhsr imagery. Computers and Electronics in Ag...

work page doi:10.1016/j.rsase.2020.100413 2020
[17]

URL http://arxiv.org/abs/2307.07930. 8 A. Data creation and pre-processing Figure A.1: Polygons in the adjacent tiles before merging and after merging. Workflow described in Section 2.9. Figure A.2: Manually generated ground truth agricultural field boundary polygons. Figure A.3: Image showing the entire image chip (size 256 by

work page arXiv

[1] [1]

doi: https: //doi.org/10.1016/ j.rse.2016.03.010

ISSN 0034-4257. doi: https: //doi.org/10.1016/ j.rse.2016.03.010. URL https://www.sciencedirect.com/science/ article/pii/S0034425716301031. Lyndon D. Estes, Su Ye, Lei Song, Boka Luo, J. Ronald Eastman, Zhenhua Meng, Qi Zhang, Dennis McRitchie, Stephanie R. Debats, Justus Muhando, Angeline H. Amukoa, Brian W. Kaloo, Jackson Makuru, Ben K. Mbatia, Isaac M....

work page 2016

[2] [2]

doi: 10.3389 /frai.2021

ISSN 2624-8212. doi: 10.3389 /frai.2021. 744863. URL https://www.frontiersin.org/articles/10.3389/ frai.2021.744863. Ashley Fernandez and Swaraj Dube. Core building blocks: Next gen geo spatial GPT application,

work page arXiv 2021

[3] [3]

Artur Gafurov

URL http://arxiv.org/abs/2310.11029. Artur Gafurov. Automated Mapping of Cropland Boundaries Using Deep Neu- ral Networks. AgriEngineering, 5(3):1568–1580, September

work page arXiv

[4] [4]

doi: 10.1016 /j.foodres.2014.03.015

ISSN 0963-9969. doi: 10.1016 /j.foodres.2014.03.015. URL https://www. sciencedirect.com/science/article/pii/S0963996914001732. Angel Garcia-Pedrero, Consuelo Gonzalo-Martin, Mario Lillo-Saavedra, and Dionisio Rodr´ıguez-Esparrag´on. The Outlining of Agricultural Plots Based on Spatiotemporal Consensus Segmentation. Remote Sensing, 10(12):1991, December

work page 2014

[5] [5]

doi: 10.3390 /rs10121991

ISSN 2072-4292. doi: 10.3390 /rs10121991. URL http: //www.mdpi.com/2072-4292/10/12/1991. Johannes Jakubik, Sujit Roy, C. E. Phillips, Paolo Fraccaro, Denys Godwin, Bianca Zadrozny, Daniela Szwarcman, Carlos Gomes, Gabby Nyirjesy, Blair Edwards, Daiki Kimura, Naomi Simumba, Linsong Chu, S. Karthik Mukkavilli, Devyani Lambhate, Kamal Das, Ranjini Bangalore,...

work page 2072

[6] [6]

Available: https://arxiv.org/abs/2310.18660

URL http://arxiv. org/abs/2310.18660. Alexander Kirillov, Eric Mintun, Nikhila Ravi, Hanzi Mao, Chloe Rolland, Laura Gustafson, Tete Xiao, Spencer Whitehead, Alexander C. Berg, Wan- Yen Lo, Piotr Doll´ar, and Ross Girshick. Segment anything,

work page arXiv

[7] [7]

Monitoring forest dynamics in Africa during 2000 – 2020 using a remotely sensed fractional tree cover dataset,

doi: 10.1080/17538947.2023.2278895. URL https://doi.org/10.1080/ 17538947.2023.2278895. Weiye Mei, Haoyu Wang, David Fouhey, Weiqi Zhou, Isabella Hinks, Josh M. Gray, Derek Van Berkel, and Meha Jain. Using deep learning and very-high- resolution imagery to map smallholder field boundaries.Remote Sensing, 14 (13),

work page doi:10.1080/17538947.2023.2278895 2023

[8] [8]

doi: 10.3390 /rs14133046

ISSN 2072-4292. doi: 10.3390 /rs14133046. Heather C North, David Pairman, and Stella E Belliss. Boundary delineation of agricultural fields in multitemporal satellite imagery. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing , 12(1): 237–251,

work page 2072

[9] [9]

doi: 10.1038 /s41467-021-24638-z

ISSN 2041-1723. doi: 10.1038 /s41467-021-24638-z. URL https://www.nature.com/ articles/s41467-021-24638-z . Publisher: Nature Publishing Group. RB Singh, Praduman Kumar, T Woodhead, et al. Smallholder farmers in india: Food security and agricultural policy. Smallholder farmers in India: Food 7 security and agricultural policy,

work page 2041

[10] [10]

doi: 10.3390/rs13040722

ISSN 2072-4292. doi: 10.3390/rs13040722. URL https://www.mdpi.com/2072-4292/13/4/

work page doi:10.3390/rs13040722 2072

[11] [11]

doi: 10.3390 / rs12071205

ISSN 2072-4292. doi: 10.3390 / rs12071205. URL https://www.mdpi.com/2072-4292/12/7/1205. Franc ¸ois Waldner and Foivos I. Diakogiannis. Deep learning on edge: Ex- tracting field boundaries from satellite images with a convolutional neu- ral network. Remote Sensing of Environment , 245:111741, August

work page 2072

[12] [12]

doi: 10.1016 /j.rse.2020.111741

ISSN 0034-4257. doi: 10.1016 /j.rse.2020.111741. URL https://www. sciencedirect.com/science/article/pii/S0034425720301115. Mo Wang, Jing Wang, Yunpeng Cui, Juan Liu, and Li Chen. Agricul- tural Field Boundary Delineation with Satellite Image Segmentation for High-Resolution Crop Mapping: A Case Study of Rice Paddy. Agron- omy, 12(10):2342, September 2022a...

work page doi:10.3390/rs14225738 2020

[13] [13]

URL https://doi.org/10.1080/17538947.2024.2353122. M. Weiss, F. Jacob, and G. Duveiller. Remote sensing for agricultural applications: A meta-review. Remote Sensing of Environment , 236: 111402,

work page doi:10.1080/17538947.2024.2353122 2024

[14] [14]

doi: https: //doi.org/10.1016/j.rse.2019

ISSN 0034-4257. doi: https: //doi.org/10.1016/j.rse.2019. 111402. URL https://www.sciencedirect.com/science/article/ pii/S0034425719304213. Qiusheng Wu and Lucas Prado Osco. samgeo: A python package for seg- menting geospatial data with the segment anything model (sam). Journal of Open Source Software, 8(89):5663,

work page doi:10.1016/j.rse.2019 2019

[15] [15]

URL https://doi.org/10.21105/joss.05663

doi: 10.21105 /joss.05663. URL https://doi.org/10.21105/joss.05663. Ruoyu Yang, Zia U. Ahmed, Urs C. Schulthess, Mustafa Kamal, and Rahul Rai. Detecting functional field units from satellite images in smallholder farming systems using a deep learning based computer vision approach: A case study from bangladesh. Remote Sensing Applications: Society and Env...

work page doi:10.21105/joss.05663

[16] [16]

doi: https: //doi.org/ 10.1016/j.rsase.2020.100413

ISSN 2352-9385. doi: https: //doi.org/ 10.1016/j.rsase.2020.100413. URL https://www.sciencedirect.com/ science/article/pii/S2352938519301831. Peng Zhang, Shougeng Hu, Weidong Li, and Chuanrong Zhang. Parcel-level mapping of crops in a smallholder agricultural area: A case of central china using single-temporal vhsr imagery. Computers and Electronics in Ag...

work page doi:10.1016/j.rsase.2020.100413 2020

[17] [17]

URL http://arxiv.org/abs/2307.07930. 8 A. Data creation and pre-processing Figure A.1: Polygons in the adjacent tiles before merging and after merging. Workflow described in Section 2.9. Figure A.2: Manually generated ground truth agricultural field boundary polygons. Figure A.3: Image showing the entire image chip (size 256 by

work page arXiv