TerraStream can be used for many forms of machine learning. Below are some examples.
Probability of Collection
The probability of collection uses the tasking geometry, satellite source, start, end, and order dates, to determine the probability of a successful collection, via a machine learning model.
The model makes predictions based on the orbital passes of the satellite source in the time range of the
tasking, the weather information, and past results, by using a linear regression algorithm to calculate overall probability.
The probability is then output in the form of a percentage.
As a user, no additional input is required beyond the information provided for your tasking order. When creating a tasking, the probability of collection will be displayed on the front end.
Cost Optimization of Taskings
The machine learning based cost optimizer groups together applicable geometries that have been requested to be captured within the same timeframe. This allows the optimally positioned satellite in a given constellation to simultaneously capture all of the requests at once.
This minimizes the number of passes required to capture the requested areas of interest. By optimizing and increasing the efficiency of tasking collection, this not only helps you as a satellite provider to minimize operational costs and increasing capture capacity, it also helps minimize fuel consumption, wear and tear of satellite components, and potentially extend the lifespan of your satellite mission.
As a satellite operator, no additional input is required, as this optimization is performed automatically on tasking requests, behind the scene.
Image Quality Prediction
Image quality prediction works by analyzing optimal satellite pass times, weather conditions, and historical data trends to estimate the probability of a successful image capture.
This takes into account the satellite orbital path, off-nadir angles, cloud cover, and past results.
A comprehensive image probability score is then output, which reflects the probability of the captured image being high quality.
As a user, no additional input is required beyond the information provided for your tasking order.
The image probability score will be displayed on the GeoPortal and Ops Center front ends, which will help you to make informed decisions for satellite imagery tasks:
The earth observation data on TerraStream allows users to perform change detection, which can be automated and improved using machine learning. This data would involve multiple orders placed over a period of time, to see any changes, and allow further investigation.
Example of this are deforestation, glacial loss from global warming, construction projects, changes in urban landscape, and changes in methane emissions.
The following animation is from a change detection project created using TerraStream data, to monitor the construction projects at the upper-middle section of the animation:
Similar to the change detection use-case, the earth observation data on TerraStream allows users to perform object detection and counting, which can be automated and improved using machine learning. This data would involve a single order placed over the area or interest, with sufficient resolution to allow the objects to be visible.
Example of this are automated counting of livestock, automated counting of aircraft or vehicle inventory, classification of pipelines for loss and accident prevention, environmental accident detection of oil wells, and drought or disease classification of crops.
The combination of complementary image resolutions and spectral bands, allow the data on TerraStream to enrich other lower cost data, often referred to as tip and cue.
Tip and cue monitoring typically begins with low-resolution optical sensors, that monitor a large area, such as Sentinel-2 imagery. This imagery is then analyzed, often via machine learning, for potential areas of interest, and is known as "Tipping". Once area for further analysis have been identified, higher resolution imagery, or imagery with different spectral bands is ordered from TerraStream, known as "Cueing".
This higher resolution data from TerraStream, allows you to enrich any lower quality data, in a cost efficient manner.