About the Client
SAGRI Co., Ltd. is a Tokyo-based agritech company, founded in 2018, that uses satellite data and AI to analyze farmland. Its AI Polygon technology automatically generates farmland polygons (digital farmland maps) from satellite imagery, including in regions where traditional maps don’t exist.
To train and evaluate AI Polygon, SAGRI needed field boundaries annotated across multiple countries. Each country brings different farmland patterns, so every new region adds boundary styles the model has to learn.
The Challenge
What Is Satellite Data Annotation for Farmland Boundary Detection?
Satellite data annotation for farmland boundary detection is the process of tracing the outline of individual fields on satellite imagery so a model learns where one parcel ends and the next begins. Annotators work across large areas of interest (AOIs), where a single dataset can hold many fields.
These polygons are the ground truth for model training and evaluation. If boundaries are drawn inconsistently, the model learns the inconsistency.
The Problem
SAGRI’s AI Polygon project needed annotation volume to grow without quality slipping.
Key challenges:
- Keeping annotation quality consistent as volume scaled across datasets
- Managing feedback cycles between QC and annotators
- Refining annotation guidelines as new farmland patterns and edge cases appeared
- Handling large AOI datasets, which took significant coordination effort
- Holding cost efficiency at large annotation volumes
What Was at Stake
Without a reliable annotation partner, SAGRI faced:
- Inconsistent boundary labels carried into model training
- Slower AI model training and evaluation cycles
- A growing coordination load on SAGRI’s own team
- Rising annotation cost as volume scaled
What Wasn’t Working
The problem wasn’t drawing polygons. It was holding one boundary standard across countries and datasets while guidelines kept evolving, at a cost that still made sense at scale. SAGRI needed a partner that could absorb detailed feedback, adapt to changing guidelines, and run large AOI datasets without adding to SAGRI’s coordination workload.
The Solution
Taskmonk set up a managed geospatial annotation workflow around SAGRI’s boundary-detection requirements. The Taskmonk team worked to SAGRI’s guidelines and adapted as those guidelines evolved.
What We Delivered
Guideline-Driven Annotation
- Annotation workflow configured to SAGRI’s boundary guidelines for each dataset
- Guideline updates applied by the annotation team as SAGRI refined its edge-case rulings
- Calibration sessions with SAGRI to shorten the annotator learning curve
Geospatial Annotation Tools
- Polygon tool for field outlines on satellite imagery
- Orthogonalise tool to snap hand-drawn polygons to clean right angles
- GeoJSON-conformant output, with GeoPackage (GPKG) import and export for pre-annotations and deliverables
QC and Feedback Loop
- Two-level workflow: L1 annotation, L2 QC review
- Maker-Checker review on submitted polygons
- Required rejection reasons, with the Rejection/Rebuttal flow keeping disagreements on record
- SAGRI’s detailed QC feedback applied by the team, shortening the gap between a guideline change and the next corrected batch
Large AOI Handling
- Multiple Batch Splitting to break large AOI imports into sequentially named batches
- Dataset Progress reporting for batch-level visibility on progress and quality
The Results
Taskmonk delivered improvements across all three priorities: quality, scale and cost.
Annotation Quality
- 98% internal accuracy against SAGRI’s calibration, held across datasets so the model trains on boundaries defined by one standard
- Realtime Guideline changes applied across the team without restarting batches, with real-time coordination on SAGRI’s suggestions
Delivery and Scale
- 125 AOIs completed at L1 with 100% completion, with images passed on to L2/QC review.
- About 10% faster turnaround, roughly 1 day saved on annotation delivery
- Large AOI projects supported without a proportional rise in coordination effort
Cost and Model Impact
- About 10% lower FTE cost, driven by timely calibration and a shorter learning curve
- Faster AI model training and evaluation cycles, because consistent, corrected annotation data arrived sooner
“TaskMonk is a reliable and flexible annotation partner that works closely with the client’s requirements, adapts well to feedback, and provides a good balance between quality, scalability, and cost effectiveness.”
Shivang Raj, Project Coordinator / Data Scientist, SAGRI Co., Ltd., Tokyo
Satellite Data Training Needs
AI Polygon is only as good as the boundaries it learns from. SAGRI needed large volumes of polygon annotations across countries, drawn to one standard, with a QC process strong enough to catch drift early.
Their datasets spanned different farmland patterns, so the workflow had to handle regional variation while keeping a single guideline. Feedback from QC had to flow back into the work quickly, and large AOI datasets had to be broken into manageable, trackable units.
Why Taskmonk for Satellite Data Annotation?
SAGRI needed a platform and team that would allow them to:
✓ Keep annotation quality consistent as volume scales across datasets
✓ Apply evolving guidelines and detailed QC feedback without slowing delivery
✓ Run large AOI datasets without heavy coordination overhead
✓ Hold cost effectiveness at large-scale annotation volumes
✓ Produce polygon data in formats that feed model training and evaluation
Taskmonk met these requirements by combining geospatial annotation tooling, a tight Maker-Checker QC loop, and a managed annotation team that owns delivery. Most platforms can draw a polygon. Holding one boundary standard across countries and months is an operations problem, and it needs the workflow, QC and team working together.
Taskmonk at a Glance
Taskmonk has processed 900M+ transactions across 7M+ labeling hours, supports 5,000+ concurrent users with 99.9% uptime on Azure, and holds a 4.6/5 rating on G2. If you’re building a geospatial or computer vision model, book a demo to see how the platform handles your imagery before you commit.
What’s Next
SAGRI’s AI Polygon work depends on a steady flow of consistent, well-calibrated boundary data. Consistent quality, faster delivery and better cost-effectiveness are what satellite data annotation looks like when guidelines, QC and the team are built around the same problem.
Building a geospatial AI program?
The quality of your training data determines everything that comes after it.
Talk to the Taskmonk team today
