Oil Platform Detection in Polarimetric SAR Images Based on Level Set Segmentation and Convolutional Neural Network

Abstract

In this paper, we proposed an oil platform detection method in polarimetric SAR images based on level set segmentation and convolutional neural network (CNN). An improved level set segmentation method is used for the extraction of regions of interest (ROIs) at first. The classic CNN model is then used for the identification of oil platforms from the extracted ROIs. In the method, the offshore strong scattering targets are coarsely detected by a thresholding segmentation of the polarimetric entropy and alpha angle parameters. Then, a circle covering the initially detected targets is obtained using a proposed circle covering algorithm. The ROIs are extracted by using level set segmentation in the initialization of the circle. Oil platforms are finally detected by using the improved LeNet-5 model. The experimental results demonstrate the effectiveness of the proposed method using multiple sets of polarimetric SAR data from different sea regions acquired by RADARSAT-2.