Low computational ship detection in Synthetic Aperture Radar satellite images using CFAR-based algorithms

Sammanfattning

Ship detection is a crucial part of reconnaissance and enforcing maritime laws, especially withthe rise of shadow vessels that turn off their Automatic Identification System (AIS) to circumvent sanctions. By utilising spaceborne Synthetic Aperture Radar (SAR), imaging is possible inall conditions due to its unique properties. With the aim to develop a program that can detectlarge ships in any kind of SAR image with as low processing time as possible, an algorithmbased on Constant False Alarm Rate (CFAR) was constructed. Three different versions of themodel were developed, the difference being how the global threshold was determined. Out ofthe three versions, one outperformed the others across the evaluation metrics. It achieved highdetection rate in addition to few false detections and low processing time. Furthermore, theprogram can create image patches of the detected ships with embedded coordinates, which maybe analysed by an expert or a trained artificial intelligence for further classification and identification of the ship. Overall, it was observed that the model performed significantly better onimages over open sea, with some diminished performance in images depicting detailed ports.To properly use the model on such images, the land mask has to be further improved. For useon images over open sea, the model can accurately detect the ships and calculate position aswell as approximate length.