Name
Abstract | ||
|---|---|---|
Aurora is a natural light in high latitude regions reflecting the interaction of atoms in solar-terrestrial space. To benefit the polar atmosphere research, this paper aims to search images of interest from the big aurora data captured by the all-sky imager (ASI). Other than employing conventional hand-crafted features, we leverage the convolutional neural network (CNN) to extract contextual features in multi-scale. Due to its strong discriminative power, false matches are effectively alleviated and the precision of visual matching is greatly improved. Specially, to conform to the aurora structure, a polar region division (PRD) scheme is explored in consideration of the imaging principle of ASI, which is more effective than the spatial pyramid matching (SPM) approach. Experimental results demonstrate that the proposed method improves the search accuracy with acceptable memory cost and efficiency. |
Year | DOI | Venue |
|---|---|---|
2018 | 10.1016/j.neucom.2017.11.059 | Neurocomputing |
Keywords | Field | DocType |
Aurora image search,Bag-of-features,Contextual CNN feature | Pattern recognition,Convolutional neural network,Bag of features,Artificial intelligence,Pyramid,Visual matching,Polar,Discriminative model,Mathematics | Journal |
Volume | ISSN | Citations |
281 | 0925-2312 | 1 |
PageRank | References | Authors |
0.35 | 33 | 4 |