期刊名称：International Journal of Signal Processing, Image Processing and Pattern Recognition
印刷版ISSN：2005-4254
出版年度：2016
卷号：9
期号：3
页码：341-354
DOI：10.14257/ijsip.2016.9.3.30
出版社：SERSC
摘要：Automatic detection for remote sensing targets is a profound challenging problem in the field of remote sensing image analysis. This paper presents a novel fast targets detection model based on kernel sparse coding with spatial bag of visual words (KSC-SBOW) in the optical remote sensing images. Before the implementation of the sliding window for targets detection, a saliency prediction model is introduced to predict the suspected targets which can extremely depress the influence of backgrounds and rationally decrease the computational cost. Following the determination of a processing sliding window by the saliency prediction model and the SIFT features extraction from the image patch, the kernel sparse coding is utilized to encode the features for a lower reconstruction error and the spatial information is added by the spatial-pyramid mapping method in the KSC-SBOW description model. Specifically, we propose the principal component analysis method based kernel orthogonal matching pursuit algorithm (KPOMP) to solve the problem of kernel sparse coding. In KPOMP, histogram intersection kernel works as the measurement kernel to more effectively capture the similarities among those Scale Invariant Feature Transform (SIFT) features and the principal component analysis method is implemented for the kernel dimensionality reduction to speed up the coding process with the guarantee of coding effectiveness. Finally, the KSC-SBOW model is combined with linear support vector machine for the targets detection. In a number of targets detection experiments, we demonstrate that the proposed model achieves outstanding performance in terms of the detection accuracy and operating rate.
关键词：Targets detection; Saliency prediction; Kernel sparse coding; Spatial bag of ; visual words; Kernel dimensionality reduction
