Computer vision
Object detection, classification, segmentation and visual feature recognition when suitable for the task.
AI technology & capabilities
The technology behind useful image intelligence.
Satvitech combines image preparation, computer vision, AI-based analysis, geospatial processing, and quality review into workflows designed around each project's objective and source data.
Example visualisationCore capability areas
Methods we can scope for your project
These are capabilities available depending on project scope — not modules of a pre-built platform. The right combination depends on the question and the data.
Image processing
Preparation, normalisation, image alignment and sensor-appropriate handling of the source data.
Geospatial processing
Spatial reference handling, feature locations, area measurements and data outputs where source metadata supports them.
Multispectral analysis
Relevant bands and derived indices — only when the source provides suitable bands and the scientific approach is appropriate.
Change analysis
Comparison of aligned, appropriate images captured at different times, with review of potential artefacts.
Automation
Repeatable processing pipelines for compatible data and agreed workflows.
Quality review
Evaluation of sample outputs, tracking of false positives and missed detections, and documented constraints.
The analysis pipeline
Five stages, each one documented
The same structure applies whether the task is a single image or a recurring workflow.
- 1 · Data assessmentSource, image type, date, resolution, metadata, availability and permitted use.
- 2 · PreprocessingPrepare, align or filter data as the task requires.
- 3 · AI / analytical processingApply a method appropriate for the target problem.
- 4 · Validation & reviewInspect results, compare with reference data where available, document limitations.
- 5 · DeliveryProvide the agreed files, visuals, reports or workflow outputs.
Imagery choices
Different imagery answers different questions
Imagery differs by sensor and acquisition characteristics. Choosing well is often the most important decision in a project.
Optical imagery
Visually intuitive and similar to a photograph.
- Watch for
- Clouds, haze, shadows, lighting and daylight-only acquisition.
Multispectral imagery
Bands beyond visible light can support certain material, vegetation or land-cover analyses.
- Watch for
- Requires the right bands and an appropriate method.
Radar imagery
Can be useful in conditions that limit optical imagery, for certain applications.
- Watch for
- Different interpretation and processing; not a drop-in replacement for optical data.
Commercial imagery
May offer higher resolution, different coverage or tasking options, depending on provider and licence.
- Watch for
- Additional cost and licence terms that define permitted use.
Open imagery & data
Suitable for some projects, particularly broader-area tasks.
- Watch for
- Coverage, resolution, revisit, timeliness and permitted uses vary by source.
Satvitech can assess open and commercial sources as appropriate for each project. We do not have exclusive access to any imagery source, and no provider named or shown on this site is a partner.
Quality and limitations
Defining “good enough” before the work starts
No imagery analysis is perfect. Each project should define what level of performance is good enough for its intended use — and what a mistake would cost.
Where appropriate, outputs include confidence scores, sample-review findings or other quality notes. A model's confidence score is not automatically a calibrated probability of being correct, so we explain how scores were reviewed. We do not promise 100% accuracy.
Reliable output depends on
- Target size compared with pixel size and source resolution
- Sensor characteristics, band availability, image date and viewing geometry
- Cloud, haze, shadows, seasonality and occlusion
- Model suitability and representative validation data
- Differences between image dates or sources
- The business meaning of a false positive versus a missed detection
False positive
Something reported that is not really there. Costly when every result triggers an action.
Missed detection
Something present that was not reported. Costly when completeness matters most.
Data sourcing and usage
The right data, used the right way
Satvitech uses appropriate third-party and open sources and/or customer-supplied imagery, depending on the project. Source selection weighs data access, licensing, resolution, date coverage, budget and the required output.
Every proposed workflow is subject to imagery availability and permitted use. If suitable data cannot be obtained for a location or date, we say so before work begins.
Access & licensing
Is the imagery obtainable, and does its licence allow the intended analysis and outputs?
Resolution
Is the detail sufficient for the smallest feature that matters?
Date coverage
Are there images for the dates — or the cadence — the question needs?
Budget
Does the cost of the data match the value of the answer?
Tools and stack
Method first, tools second
Tools, frameworks and processing environments are chosen per project to suit the method, the data volume and how results will be delivered. We would rather describe what a workflow does than show a wall of vendor logos.
If your project has requirements for particular formats, software or deployment environments, include them in your enquiry and we will take them into account during scoping.
Discuss the technical requirements of your project.
Share your objective, any data or format constraints, and how outputs need to be delivered.