Deep Blue Observer Maritime Domain Awareness Joint Challenge

Context

Current maritime domain awareness is hindered by unruly, high-volume, heterogeneous data streams that overwhelm analysts. This is further compounded by classification barriers that limit information-sharing with allies and partners.

Given the density of traffic through the Singapore straits and adjacent sea lanes, we are seeking an integrated intelligence layer capable of ingesting, correlating, and automating threat detection of illicit vessels across Singapore's waters and its maritime approaches.

What We Are Looking For

• Solutions that automate fusion across multi-INT sources such as Geospatial Intelligence (GEOINT), imagery, Automatic Identification System (AIS), Signals Intelligence (SIGINT), and open-source maritime data into a single operating picture around Singapore waters (up to 1,600nm from Singapore to a water depth of 200m below mean sea level).
• Ingested intelligence must be processed and transmitted in near-real time manner to support compressed decision cycles.
• Solutions which are platform- and sensor-agnostic.
• Relevant proprietary datasets to detect adversarial vessels across three categories: (i) <25m, (ii) 25m-50m and (iii) >50m.

Key Metrics

• Prediction Accuracy.
• Explainability.
• Granularity.
• Computational Cost.
• Latency.
  

Your Technology Solution Should Demonstrate

• Accurate prediction of illicit vessels (piracy, smuggling, terrorism, etc.), anomalies (AIS spoofing, irregular deviations from expected route, suspicious vessel behaviour etc.), and other maritime threats in Areas of Interest.
• Low computational cost to enable edge deployments.
• Platform and sensor-agnostic interfaces.
• Solution to be web-based or via an Application Programming Interface (API) .
  

What A Trial Looks Like

• Live demonstration of shortlisted solutions leveraging maritime activity over the last 48 hours. (i.e. 48 hours prior to scheduled demo).
• Automated detection, including tasking one sensor based on another sensor's detection.
• Detections returned with a likelihood score rather than a yes/no alert, showing which sources drove them.
• Forward projection of likely vessel behaviour, over a forecast window the applicant defines and justifies.
• Evaluation of latency, accuracy, and operator workload reduction.
  

What We Are Not Looking For

• Solutions that rely solely on historical data to provide predictions.
• Solutions requiring proprietary platforms or modifications to native software stacks.
• Single-source or siloed intelligence feeds.

Follow-On Implementation Cost

• Man-effort cost for customisation and deployment.
• Licence cost (if applicable).