A False-Alarm-Controllable Modified AdaBoost Wake Detection Method Using SAR Images

Abstract

A false-alarm-controllable modified AdaBoost based method is proposed for detecting ship wake from sea clutter in synthetic aperture radar (SAR) images. It reformulates the wake detection problem as a binary classification task in the multi-feature space. The update strategy of the sample weights in the original AdaBoost is modified for wake detection. First, a detection result confidence factor is designed to deal with class imbalance between sea clutter and ship wake; then, the AdaBoost is further modified as a false alarm rate (FAR) controllable detector by introducing penalty parameters to adjust weights update strategies for the sea clutter. Meanwhile, the multi-feature space is spanned by a novel frequency peak height ratio (FPHA) feature and four salient features. FPHA is proposed to enhance separation between the wake and sea clutter, which is computed from the amplitude spectrum peak of the image after Fourier transform. Experimental results show that the proposed detector can tackle the imbalanced data problem and flexibly control FAR by adjusting penalty parameters. Moreover, improved detection probability is also achieved compared with existing methods.