Skip to main content
Remote sensing teams annotate satellite, aerial, and drone imagery for land use classification, mapping, infrastructure monitoring, and environmental analysis. Avala provides polygon, segmentation, and classification tools optimized for overhead imagery, along with dataset management features for organizing large-scale image collections across geographies and time periods.

​
Data Types

SourceAvala Data TypeTypical Annotation
Satellite imageryImagePolygons, segmentation masks, classification
Aerial photographyImageBounding boxes, polygons
Drone imageryImage, VideoBounding boxes, polygons, segmentation
OrthophotosImageSegmentation masks, classification
Time-series compositesImageClassification, change detection annotations

​
Common Tasks

​
Land Use Classification

Segment satellite and aerial images into land cover categories: urban, forest, water, agriculture, barren land, wetland, and other terrain types. Use the segmentation tool for pixel-level classification across the full image, or polygon annotation for region-level labeling. Label taxonomy example: 
{
  "labels": [
    { "name": "urban", "color": "#FF4444" },
    { "name": "forest", "color": "#22AA22" },
    { "name": "water", "color": "#4488FF" },
    { "name": "agriculture", "color": "#AADD44" },
    { "name": "barren", "color": "#BBAA88" },
    { "name": "wetland", "color": "#66BBAA" }
  ]
}
For land use classification, define your categories based on the resolution of your imagery. High-resolution satellite (0.3-0.5 m/pixel) supports fine-grained categories like “residential” vs. “commercial.” Lower resolution (10-30 m/pixel) works better with broader categories like “urban” vs. “rural.”

​
Building Footprint Extraction

Trace polygon boundaries around buildings for mapping, urban planning, and change detection. The polygon tool supports:
  • Vertex-level editing for precise rooftop boundaries
  • Edge snapping for rectilinear buildings
  • Copy-and-adjust for repetitive structures (e.g., row houses)
Building footprint datasets are used in population estimation, disaster response mapping, and construction monitoring.

​
Vehicle and Object Counting

Detect and count vehicles, ships, aircraft, and other objects in overhead imagery using bounding box annotations. For dense scenes (parking lots, ports, airports), bounding boxes provide fast annotation with sufficient precision for counting and detection models.

​
Infrastructure Monitoring

Annotate infrastructure elements — roads, bridges, power lines, solar panels, pipelines — for condition assessment and change detection. Common approaches include:
InfrastructureAnnotation TypeAttributes
RoadsPolylinesSurface type, condition, width
BuildingsPolygonsRoof type, damage level, construction status
Power linesPolylinesSpan type, tower presence
Solar panelsPolygonsPanel count, orientation
Water bodiesPolygonsType (river, lake, reservoir), boundaries

​
Change Detection

Compare images from different dates to identify changes: new construction, deforestation, flood extent, crop growth. Organize temporal image pairs in the same dataset and annotate changes with classification labels and polygon boundaries. Use metadata fields to tag images with collection date and location: 
metadata.collection_date = "2026-01-15"
metadata.region = "san-francisco-bay"
metadata.satellite = "worldview-3"
Then filter with the query language to find specific temporal windows: 
metadata.region = "san-francisco-bay" AND metadata.collection_date = "2026-01-15"

​
Dataset Organization

Satellite and aerial imagery datasets tend to be large — thousands to millions of images across geographies and time periods. Effective organization is critical.

​
Organization Strategies

StrategyWhen to UseExample
By regionMulti-geography projectsnorth-america, europe, southeast-asia
By collection dateTemporal analysisq1-2026, q2-2026
By resolutionMixed-resolution sourceshigh-res-30cm, medium-res-10m
By taskDifferent annotation goalsbuilding-footprints, land-cover, vehicle-counting

​
Using Slices

Slices create virtual subsets without duplicating images:
  • Create a training slice with 80% of images and a validation slice with 20%
  • Create slices by geographic region for region-specific model evaluation
  • Create a difficult-cases slice for images that annotators frequently get wrong

​
Cloud Storage Integration

For large satellite imagery collections, use cloud storage integration to connect your S3 or GCS bucket directly. Avala reads images from your bucket without requiring a separate upload step.
Cloud storage is recommended for satellite imagery datasets over 10,000 images. It avoids the upload bottleneck and keeps your data in your own storage with your encryption and retention policies.

​
Avala Features Used

FeaturePurposeLearn More
Polygon annotationBuilding footprints, infrastructure boundariesPolygon Tool
Segmentation annotationPixel-level land cover classificationSegmentation Tool
Bounding box annotationVehicle and object detectionBounding Box Tool
Polyline annotationRoads, power lines, and linear featuresPolyline Tool
ClassificationScene-level land type and condition labelsClassification Tool
AutoTagSimilarity-based grouping for scene discoveryAutoTag
Dataset managementOrganize imagery by region, date, and sourceManaging Datasets
SlicesCreate training/validation splits and region subsetsSlices API
Cloud storageConnect S3 or GCS for large imagery collectionsCloud Storage
Quality controlMulti-stage review for mapping-grade accuracyQuality Control

​
Example Pipeline

Satellite / aerial imagery collection
  -> Store in S3 or GCS bucket
  -> Connect to Avala via cloud storage integration
  -> Create dataset, attach metadata (region, date, satellite)
  -> Create slices for training/validation splits
  -> Create annotation project (segmentation for land cover, polygons for buildings)
  -> Annotators label imagery with review stages
  -> Export in COCO or GeoJSON-compatible format
  -> Train remote sensing model
  -> Use model predictions for next round of auto-labeling

​
Getting Started

1

Prepare your imagery

Convert images to JPEG or PNG format. For georeferenced data, keep the coordinate metadata in sidecar files or your GIS system — Avala handles the image pixels.
2

Upload or connect storage

For small datasets, upload directly. For large collections, connect your S3 or GCS bucket via cloud storage integration.
3

Organize with metadata and slices

Attach metadata (region, date, source satellite) to items. Create slices for training/validation splits and geographic subsets.
4

Create annotation project

Define your label taxonomy (land cover classes, building types, infrastructure categories). Choose the annotation type that fits your task.
5

Annotate and review

Distribute work across your team with work batches. Use multi-stage review for mapping-grade accuracy requirements.
6

Export and train

Export annotations and integrate with your remote sensing training pipeline via the API or SDK.

​
Next Steps

Polygon Tool

Precision boundary tracing for building footprints and infrastructure.

Segmentation Tool

Pixel-level classification for land cover and terrain mapping.

Cloud Storage

Connect your S3 or GCS bucket for large imagery collections.

Best Practices

Dataset organization, API usage, and annotation workflow optimization tips.