Abstract
This paper proposes an innovative Adaptive Component Selection-Based Discriminative Model (ACSDM) for object detection in high-resolution synthetic aperture radar (SAR) imagery. In order to explore the structural relationships between the target and the components, a multi-scale detector consisting of a root filter and several part filters is established, using Histogram of Oriented Gradient (HOG) features to describe the object from different resolutions. To make the detected components of practical significance, the size and anchor position of each component are determined through statistical methods. When training the root model and the corresponding part models, manual annotation is adopted to label the target in the training set. Besides, a penalty factor is introduced to compensate information loss in preprocessing. In the detection stage, the Small Area-Based Non-Maximum Suppression (SANMS) method is utilised for filtering out duplicate results. In the experiments, the aeroplanes in TerraSAR-X SAR images are detected by the ACSDM algorithm and different comparative methods. The results indicate that the proposed method has a lower false alarm rate and can detect the components accurately.
Highlights
Synthetic aperture radar (SAR) is a microwave imaging sensor proposed by Carl Wiley in 1951 [1,2] to obtain high-resolution synthetic aperture radar (SAR) images
In the Constant False Alarm Rate (CFAR)-based algorithm, it is calculated that the detection precision P = 73.2%, the false alarm rate Pf a = 26.8%, and the recall rate R = 93.2%
It is clear that the method based on CFAR for target detection in high-resolution SAR imagery is not reliable enough
Summary
Synthetic aperture radar (SAR) is a microwave imaging sensor proposed by Carl Wiley in 1951 [1,2] to obtain high-resolution SAR images. Containing rich scattering and polarization information, SAR imagery is widely used and can be formed regardless of time or weather [3]. Target detection methods in this field can be divided into three categories: (1) the scattering center model-based detection algorithm [4]; (2) the statistical feature-based detection algorithm [5]; (3) the detection algorithm which introduces classical optical detection methods [6]. In the scattering center model-based detection algorithm, the model is first set-up with parameters determination; the intensity, location, and structure information of the scattering center are obtained to set a threshold for target detection. Setting different parameters and resolutions, these models can be converted to each other
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have