Feature Synthesis, and Target Discrimination and Classification for High-Resolution SAR Images

• Main Authors: Sina Hadipour Lakmeh Sari, 西那
• Other Authors: CHANG, YANG-LANG
• Format: Others
• Language: en_US
• Published: 2019
• Online Access: http://ndltd.ncl.edu.tw/handle/gvu95c

碩士 === 國立臺北科技大學 === 電機工程系所 === 107 === In this study existing and new features of high-resolution synthetic aperture radar (SAR) target images are investigated and these features are combined to create a high dimensional feature space to be used by classifiers. There are three main stages in this process: preprocessing, feature extraction, and classification. For preprocessing (also known as prescreening), a refined Lee filter was used. Refined Lee filter uses directional masks to remove additive and multiplicative noise and at the same time preserve sharp edges. In the second stage, which is the main purpose of this study, different categories of features are extracted from the target images. These categories are: spatial boundary features (5 features), contrast features (3 features), standard deviation (1 feature), and CFAR features (17 features). There is a total of 28 features used in this study. Finally, the 28-dimensional feature vectors are fed into different classifiers for training. Experimental results on Moving and Stationary Target Acquisition and Recognition (MSTAR) dataset using SVM, KNN, and Random Forest classification methods showed overall detection accuracy of 96.04%, 93.33%, and 94.13% respectively.