Rain-Contaminated Region Segmentation of X-Band Marine Radar Images With an Ensemble of SegNets

The presence of rain may blur surface wave signatures and cause additional radar backscatter, which negatively affects the performance of ocean remote sensing applications (e.g., ocean surface wind and wave parameter measurement) using X-band marine radars. In this article, a novel end-to-end model is developed to detect and locate rain-contaminated pixels in X-band marine radar images based on a type of deep neural network called SegNet, which is able to segment rain-contaminated regions by classifying each pixel into three classes: rain-free, rain-contaminated, and wind-dominated rain cases. Shipborne marine radar images collected during a sea trial on the East Coast of Canada are first preprocessed and then utilized to train an ensemble of SegNet-based networks. The final classification result of each pixel will be the class chosen by most individual networks. Testing results using images obtained from both shipborne and shore-based marine radar systems manifest that the proposed model effectively segment between rain-free, rain-contaminated, and wind-dominated rain regions, with a pixel classification accuracy of 94.6% and 90.4% for Decca and Koden radar images, respectively.