Harbor Detection in SAR Images Based on Multidirectional One-Dimensional Scanning

In SAR image target detection, harbor detection can help the detection of harbor targets and maritime traffic planning. In this paper, we propose a harbor detection method of SAR images based on multidirectional one-dimensional scanning. Take the candidate points along the coastline and the multidirectional one-dimensional scanning is performed. Using the distribution characteristics of land, sea and dock in the one-dimensional vector, training a convolutional neural network to classify the candidate points into harbor and non-harbor feature points. Then we get the harbor feature points map reflecting the distribution of harbors. The Sentinel-1 spaceborne SAR images covering a coastal region are used to verify the proposed method. The experimental results show the effectiveness and accuracy of the proposed method.