GeoAI Object Extraction

Extract objects from high-resolution satellite, aerial, or UAV imagery using foundation models, specialized tree-crown models, pretrained detection models, or project-trained custom models.

Remote Sensing CoreEnd-User DocumentationTheory + Formula + Parameters

1. What This Feature Does

Extract objects from high-resolution satellite, aerial, or UAV imagery using foundation models, specialized tree-crown models, pretrained detection models, or project-trained custom models.

Cloud workflow: choose data → configure the scientific method → run → review the resulting layer/report. No programming is required.

2. Recommended Workflow

Choose the high-resolution raster. ↓ Select an extraction model. ↓ Choose target object / text prompt / project model. ↓ Set confidence, tiling and geometry options. ↓ Run extraction. ↓ Review and edit the resulting objects before operational use.

3. Theory, Methods & Equations

SAMGeo3 — Zero-Shot Text Prompt

Uses a segmentation foundation model guided by a text prompt to propose object masks without custom training for the target scene.

Mask = SegmentationModel(Image, Prompt)

Best used when: rapid first-pass extraction is needed for a concept that can be described by text.

detectree2 — Individual Tree Crown

Uses a Mask R-CNN-family instance-segmentation workflow specialized for delineating individual tree crowns in high-resolution imagery.

IoU = |Prediction ∩ Reference| / |Prediction ∪ Reference|

Best used when: individual tree crown polygons are the target.

YOLO26 — Generic Pretrained

Detects/segments classes learned during generic pretraining; confidence filtering and non-maximum suppression remove weak/duplicate detections.

IoU = A_intersection/A_union

Best used when: the target resembles classes represented by the pretrained model.

Custom Project Model

Runs a compatible model previously trained for the user's own target classes.

Best used when: domain-specific imagery requires reliable, repeatable extraction.

4. Input Data

Input	Type	Requirement	Description
High-Resolution Raster	raster	Required	High-resolution satellite, aerial, or UAV raster input.

5. Parameters Available in the Application

Parameter	Default	Choices / Range	How to Use It
GeoAI Algorithm	—	SAMGeo3 — Zero-Shot Text Prompt, detectree2 — Individual Tree Crown, YOLO26 — Generic Pretrained Detection/Segmentation, YOLO — Custom GISCollab Model
Target Object	building	Building, Car / Vehicle, Ship / Vessel, Oil Palm, Individual Tree, Solar Panel, Custom Text / Class	Shown for: samgeo3_text, yolo26_pretrained, yolo_custom
Custom Text Prompt / Class Name	—	—	For SAMGeo3 this is a text prompt. For YOLO this is a class name filter. Shown for: custom
Output Geometry	polygon	Polygon / Area, Point / Centroid, Line / Boundary	Polygon keeps object footprints, Point returns representative centroids, and Line returns object boundaries.
RGB Bands	[1, 2, 3]	—	Three 1-based bands in Red, Green, Blue order used to create an 8-bit model input image.
Inference Tile Size	1024	256 to 2048
Tile Overlap	128	0 to 512
Minimum Object Size (pixels)	25	1 to —	Shown for: samgeo3_text
SAMGeo3 Backend	transformers	Transformers (CPU / GPU), Meta (NVIDIA GPU recommended)	Shown for: samgeo3_text
Tree Crown Model	default	Default, Paracou, Sepilok, Danum	Shown for: detectree2_tree
Confidence Threshold	0.35	0.0 to 1.0	Shown for: detectree2_tree, yolo26_pretrained, yolo_custom
IoU / Duplicate Removal Threshold	0.5	0.0 to 1.0	Shown for: detectree2_tree, yolo26_pretrained, yolo_custom
Custom Model Reference	—	—	cloud storage/local model reference, for example {"key":"models/1/2/geoai/building_best.pt"}. Shown for: yolo_custom
Device	auto	Auto, CPU, CUDA GPU, Apple MPS
Maximum Output Objects	100000	1 to 1000000

6. Output & Interpretation

The result should be interpreted according to the selected method and the physical meaning of the input data. Preserve source units, coordinate reference information, NoData meaning, acquisition date, and preprocessing level when comparing results.

Scientific interpretation: an algorithm can produce a numerically valid result even when the input data are unsuitable. Always check masks, units, sensor characteristics, spatial resolution and reference data.

Important: Foundation and pretrained models should be reviewed carefully on geospatial imagery because image scale, viewing angle and domain can differ from natural-image training data.

7. Best Practices

Use analysis-ready inputs and remove invalid/cloud/noise artifacts that are not part of the target phenomenon.
Choose parameters from the physical scale of the data, not only from visual appearance.
Keep categorical and continuous rasters conceptually separate when selecting interpolation/resampling methods.
Compare the result with the source image and independent reference information.
Document the settings used when results will be compared across dates, sensors or study areas.

← ROI Analysis GeoAI Semantic Classification →