A Lightweight and Fast Method for Water and Airport Segmentation in SAR Images

Abstract:
In large scene SAR images, the extraction of regions of interest is crucial for improving the accuracy of target detection. Airports and water bodies within these scenes often contain aircraft and ships, which are primary targets for detection. SAR images exhibit similar scattering characteristics for airport runways and water bodies. In light of this, our paper presents a lightweight and fast segmentation method applicable to both airport and water regions(LFSAW). Firstly, we construct a lightweight and efficient concatenate module (LECM) for rapid feature extraction. Subsequently, we create a specialized module for SAR images that extracts edge information (DIEM) from ground truth and utilizes it for supervision while simultaneously enhancing the segmentation performance by extracting detailed information. Finally, we employ a multi-task loss function to jointly supervise the edge detail and segmentation information. Furthermore, to facilitate simultaneous segmentation of airport and water regions, we build a dataset comprising 7931 image slices designed specifically for airport-water body segmentation tasks.