Sea-surface floating small target detection based on feature compression

Abstract

Owing to characteristics of sea clutter and diversity of floating small targets, it is significant to extract features to jointly detect targets. Due to the complexity of detectors in high-dimensional (HD) space, feature compression is an important procedure in the design of detector. Besides, considering that the capacities of detecting target about extracted features are varied with different datasets, feature selection is supposed to be an effective method. Here, it is found that building a feature-compression matrix can realise that mapping the feature vectors in HD space into low-dimensional space, where the matrix is built efficiently by using the results of feature selection. Whereas information about targets which is used in building feature-compression matrix is unknown, a training sample generator which can emulate the fundamental state of targets to help to build a feature-compression matrix is proposed. Finally, a one-class classifier about the feature vector which has been compressed is provided with using a new 3D convexhull learning algorithm. The experiment results on the IPIX datasets show that the proposed detector attains better detection performance than several existing detectors.

Keywords

• radar clutter
• feature extraction
• marine radar
• learning (artificial intelligence)
• object detection
• radar detection
• matrix algebra
• vectors
• feature selection
• feature-compression matrix
• sea clutter
• feature selection
• low-dimensional space
• feature vector extraction
• sea-surface floating small target detection
• hd space detectors
• high-dimensional space detectors
• training sample generator
• 3d convexhull learning algorithm
• ipix datasets