Validating Targets Detected by SAR Ship Detection Engines

• Published in: Canadian journal of remote sensing Vol. 43; no. 5; pp. 451 - 454
• Main Authors: Sandirasegaram, Nicholas, Vachon, Paris W.
• Format: Journal Article
• Language: English
• Published: Taylor & Francis 03.09.2017; Taylor & Francis Group
• ISSN: 0703-8992; 1712-7971
• DOI: 10.1080/07038992.2017.1342204

Validating detected targets by ship detection engines without ground-truth data is a very challenging task and this is addressed in this article. Targets detected in synthetic aperture radar (SAR) imagery are validated against automatic identification system (AIS) messages and using visual assessment procedures. Targets detected by the OceanSuite ship detection engine were used as the baseline for validation purposes. OceanSuite has been tuned to minimize the false alarm rate while providing a good overall detection rate. RADARSAT-2 images are often used for ship detection; maritime satellite surveillance radar mode images are considered in this article. An appropriate (spatial and temporal) AIS message is first selected, then the SAR ship signature location is predicted; the predicted ship signature location may be automatically associated with the OceanSuite detections based on proximity. Finally, all detections are validated through visual assessment. Key outcomes of this work are the assessment criteria for declaration as a ship target, tools for automatic validation of SAR detections against AIS-reported targets, and visual assessment results. The RADARSAT-2 images used in this work were collected near Vancouver Island and the Canary Islands. OceanSuite demonstrated a 78% accuracy for AIS-reported targets and over 80% of the validated targets were declared by OceanSuite. Visual assessment showed that a few false targets were declared by the OceanSuite ship detection engine.