Skylight uses a type of machine learning called “Computer Vision” to detect vessels from very large, remote sensed data (imagery) to detect vessels.

While it's possible to manually analyze images to detect vessels, machine learning is able to scan huge images to quickly identify objects that are likely to be vessels. 

In general, satellite imagery is a powerful tool to help detect vessels that may not be transmitting AIS location data. While the imagery and capabilities do not mirror that of Hollywood, the ability to quickly scan huge areas represents a step forward in maritime domain awareness.

Skylight processes several types of publicly available imagery, most of which has never been used to detect vessels or provided at no cost.

For in-depth information about the computer vision machine learning models see: 

• SatlasPretrain: A Large-Scale Dataset for Remote Sensing Image Understanding
• Satellite Imagery and AI: A New Era in Ocean Conservation, from Research to Deployment and Impact

What is remote sensing?

Remote Sensing is the process to detect objects through imaging, radar, and other signals commonly gathered from aircraft or satellites. It involves acquiring data about the Earth's surface without direct physical contact. 

This is achieved through the use of various sensors that can detect and measure electromagnetic radiation (such as visible light, infrared, microwave, or radio waves) reflected or emitted from the target area.

Chances are you have already come across some type of remote sensing. It has a wide array of applications across many sectors including agriculture, land use mapping and monitoring, disaster management, monitoring ocean characteristics, climate monitoring, urban planning, weather forecasting, forest mapping, water management, and mining.

Within the maritime space, a range of satellite remote sensors support Monitoring Control and Surveillance efforts.

This section was developed with materials from the Joint Analytical Cell. For more information, please contact the Joint Analytical Cell: jac-coord@tm-tracking.org.

Remote sensors for maritime surveillance

Skylight applies machine learning to a variety of remote sensor type data/imagery.  These include:

• Satellite Radar (Synthetic Aperture Radar, SAR)
• Optical Imagery (Electro Optical Imagery, EO)
• Night Lights (Visible Infrared Imaging Radiometer Suite, VIIRS)
• Radio Frequency (RF)

Each type and individual censor has strengths and weaknesses depending on circumstances and use case.

The following articles will introduce each one in more detail, as well as the satellites involved. 

Are vessel detections available real time?

No. There is a delay from the time the satellite captures to the time vessel detections are available from Skylight. The vast majority of this delay is the time to send the data from a satellite back to earth.  This delay is called Latency.

The latency is different for each satellite and geographies. Some vessel detections may be available in as little as an hour. Others averages between 3-7 hours. Skylight cannot control this latency. 

Satellites remain a complex, challenging endeavor. Its always possible a transmission may fail resulting in delayed or absent data. Satellites can experience issues causing gaps in data collection.

Interpreting vessel detections

Machine learning is tool to quickly process huge images and quickly identify likely vessels. However, false positives (errors) do occur.

It is critical that analysts evaluate individual vessel detections based on their local knowledge.

What size vessels can be detected? 

It depends. Small wooden boats are the most difficult to detect. 

The ability to detect a vessel from satellite imagery depends on numerous factors. Weather conditions, vessel material, vessel size, sea state and other factors can all impact results.

Each article on the various sensors will provide more detail on strengths and weaknesses. 
 

Key terminology

Detection - The possible presence of a vessel

Frame - The area where the satellite was looking. Also known as a collection, an acquisition, an image, or footprint.

Tile - a smaller portion of a frame

Swath - The path along which a satellite images/collects data.

Revisit rate - The frequency a satellite, or constellation (group) of satellites images the same area. A five day revisit rate means the satellite will capture an image of the same area every 5 days. Revisit rate is also known as temporal resolution.

Coverage - All areas a satellite, or constellation (group) of satellites images

Resolution - Spatial Resolution is the detail within an image, commonly based on the number of pixels.

Higher resolution imagery indicates more pixels in a given area. For example, 1 meter resolution means that each pixel represents a 1x1 meter area.