Size Invariant Ship Detection using SAR imagery

Abstract

Exploration and expansion in maritime commerce over the years have fostered an increase in the number of ships in the marine regions. The proposed system can constantly monitor ships in ocean waters irrespective of weather conditions and relay valuable data for analysis. This method uses a combination of image processing and shape files on Synthetic Aperture Radar (SAR) images. Most land water discrimination techniques that rely solely on image processing fail to mask out the smaller islands with dimensions similar to a ship. Hence, the image processing is aided by reference shape files. A vector mask is applied to avoid false detections on land. The radiometric correction is done to provide an image in which the pixel values are directly associated with the radar backscatter of the scene. This makes further pre-screening easier and more accurate. Ships are detected with a Constant False Alarm Rate (CFAR) detector using the Adaptive Thresholding algorithm. Object Discrimination is used to limit false detection based on object dimension. Further, the LeNet-5 image classifier algorithm is applied to increase accuracy. Finally, a vector file is generated with the XY coordinates, latitude and longitude values, length and width of the ship, ship size ratio, and ship category.