Ship detection from X-band SAR images using multi-level multi-scale feature pyramid network

Ship detection is one of the essential technique for marine monitoring. Moreover, ship detection using SAR images is necessary to detect ships irrelevant to climate condition and data acquisition time. In SAR images, ships are usually occupy very small portion, so that detecting ships in SAR image is not simple. There are various SAR ship detection studies using deep learning and many of them achieved encouraging results. However, those studies mainly consider its model architecture and didn't give much attention to input SAR images itself. In this study, we will extract features from multi-level multi-scale feature pyramid network and consider SAR images' conditions which can affect detecting performance such as speckle noise and wave texture etc. This study will contribute for detecting ships with various sizes in SAR images more accurately and improve SAR images' usage in ship detection field.