SAR TARGET RECOGNITION USING FEATURE EXTRACTION OF MULTI-SCALE MONOGENIC COMPONENTS

In this paper, we combine the monogenic signal with the histogram of oriented gradient (HOG) feature, and propose a new low-level feature: Monogenic-HOG (MG-HOG); then use the bag of words (BoW) model to construct the middle-level feature. Most of the existed features are directly extracted from grayscale synthetic aperture radar (SAR) images. As an extended representation of analytic signals in high-dimensional space, the monogenic signal has excellent characteristics such as rotation and scale invariance in the potential extraction process, and it is also valuable in the image field. However, in the past related work, it is mostly used directly as the feature representation of the image. After experimental validation on the moving and stationary target acquisition and recognition (MSTAR) data set, the proposed feature extraction method has better performance than a number of existed methods.