Ship classification in TerraSAR-X SAR images based on classifier combination

Abstract

Ship classification is an important step in maritime surveillance utilizing synthetic aperture radar images. In this paper, we focus on the classifier architecture. The paper investigates three individual classifiers, i.e., the K nearest neighbor classifier, the Bayes classifier, and the backpropagation neural network classifier from the viewpoint of discrimination measurements firstly. Then, we propose a SVM combination strategy to fuse the results of individual classifiers. Extensive experiments conducted on the TerraSAR-X SAR images validate the effectiveness of the proposed method.

Maritime targets classification based on CNN using Gaofen‐3 SAR images

“…Besides, there are many studies applying machine‐learning methods like supportive vector machine (SVM) to target classification [4, 5]. As the SVM is essentially a binary classifier, combining multiple classifiers in image classification is a must, which increases the training cost.…”

Ship Classification in High-Resolution SAR Images Using Deep Learning of Small Datasets

“…Due to their all-weather, all-day, and high-resolution advantages, synthetic aperture radar (SAR) images have recently been used for ship classification in marine surveillance. There are several satellites that have provided high-resolution SAR images since 2007, such as ASI’s COSMO-SkyMed, DLR’s TerraSAR-X, Japan’s ALOS-2, and China’s Gaofen-3, These high-resolution SAR images provide a resolution greater than 3 m that contain rich information about the targets, such as the geometry of ships, which makes discriminating different types of ships possible [ 1 , 2 , 3 , 4 ].…”

“…The methods used for ship classification with SAR images mainly focus on feature selection and optimized classifier techniques [ 1 , 2 , 3 , 4 , 5 , 6 , 7 , 8 , 9 , 10 , 11 , 12 , 13 ]. Currently, commonly used features are (1) geometric features, such as ship length, ratio of length to width, distribution of scattering centers, covariance coefficient, contour features [ 11 ], and ship scale; and (2) scattering features, such as 2D comb features [ 7 ], local radar cross section (RCS) density [ 1 ], permanent symmetric scatterers [ 12 ], and polarimetric characteristics [ 13 ].…”

Ship Classification and Detection Based on CNN Using GF-3 SAR Images

“…Experiments based on the GF-3 dataset demonstrate the merits of the proposed methods for marine target classification and detection.resolutions and covering different regions in the world have been obtained. Up to now, researches on marine target classification in SAR images are mainly focused on large ships with distinctive features such as oil tankers, container ships, and cargo ships [2][3][4][5][6]. The scattering characteristics of the three kinds of ships have been fully exploited by some initial works [2,6].…”

“…While these feature-based classifiers can achieve high performance, the features have to be carefully designed especially when dealing with a wide variety of targets. There are also some works focused on combining the complementary benefits of traditional machine learning classifiers [4,8]. However, the classifier-combination strategy increases the computational complexity as it applies the classifiers one by one.Different from the carefully designed feature-based methods mentioned above, the CNN based methods can extract the deep features of targets automatically, which have made great progress in object classification and recognition in recent years.…”