Title: SAR ATR based on displacement- and rotation-insensitive CNN
Abstract:Among many synthetic aperture radar (SAR) automatic target recognition (ATR) algorithms, convolutional neural network (CNN)-based algorithms are the commonly used methods. However, most previous SAR A...Among many synthetic aperture radar (SAR) automatic target recognition (ATR) algorithms, convolutional neural network (CNN)-based algorithms are the commonly used methods. However, most previous SAR ATR studies assume that the precise location (and heading direction) of a target is (are) known and image is not suffering from translations, which are not always true in realistic applications. In this letter, a modern CNN model is trained by samples with no rotation and displacement, and is evaluated on the dataset with rotation and displacement. The results show that the classification accuracy is very low when the target's displacement or rotation angle is different from the pre-assumed value in the training dataset. To overcome this problem, a displacement- and rotation-insensitive deep CNN is trained by augmented dataset. The proposed method is evaluated on moving and stationary target acquisition and recognition (MSTAR) dataset. It proves that our proposed method could achieve high accuracy in all three subsets which have different displacement and rotation settings.Read More
Publication Year: 2016
Publication Date: 2016-06-20
Language: en
Type: article
Indexed In: ['crossref']
Access and Citation
Cited By Count: 73
AI Researcher Chatbot
Get quick answers to your questions about the article from our AI researcher chatbot
Title: $SAR ATR based on displacement- and rotation-insensitive CNN
Abstract: Among many synthetic aperture radar (SAR) automatic target recognition (ATR) algorithms, convolutional neural network (CNN)-based algorithms are the commonly used methods. However, most previous SAR ATR studies assume that the precise location (and heading direction) of a target is (are) known and image is not suffering from translations, which are not always true in realistic applications. In this letter, a modern CNN model is trained by samples with no rotation and displacement, and is evaluated on the dataset with rotation and displacement. The results show that the classification accuracy is very low when the target's displacement or rotation angle is different from the pre-assumed value in the training dataset. To overcome this problem, a displacement- and rotation-insensitive deep CNN is trained by augmented dataset. The proposed method is evaluated on moving and stationary target acquisition and recognition (MSTAR) dataset. It proves that our proposed method could achieve high accuracy in all three subsets which have different displacement and rotation settings.