Multi-class geospatial object detection and geographic image classification based on collection of part detectors

• Published in: ISPRS journal of photogrammetry and remote sensing Vol. 98; pp. 119 - 132
• Main Authors: Cheng, Gong, Han, Junwei, Zhou, Peicheng, Guo, Lei
• Format: Journal Article
• Language: English
• Published: Amsterdam Elsevier B.V 01-12-2014; Elsevier
• Subjects: Animal, plant and microbial ecology; Applied geophysics; Biological and medical sciences; Collection; Collection of part detectors (COPD); Detectors; Earth sciences; Earth, ocean, space; Exact sciences and technology; Fundamental and applied biological sciences. Psychology; General aspects. Techniques; Geographic image classification; Geospatial object detection; Image acquisition; Image classification; Image detection; Internal geophysics; Part-based model; Remote sensing; Remote sensing images; Support vector machines; Teledetection and vegetation maps; Very-high-resolution (VHR); Visual; Geospatial object detection; Remote sensing images; Part-based model; Geographic image classification; Very-high-resolution (VHR); Collection of part detectors (COPD); models; detection; orientation; seeds; rotation; remote sensing; technology; classification; Image; Spatial distribution; solution; collections; discoveries; spatial resolution; spatial variations; data bases; frame structure; high resolution
• ISSN: 0924-2716; 1872-8235
• DOI: 10.1016/j.isprsjprs.2014.10.002

The rapid development of remote sensing technology has facilitated us the acquisition of remote sensing images with higher and higher spatial resolution, but how to automatically understand the image contents is still a big challenge. In this paper, we develop a practical and rotation-invariant framework for multi-class geospatial object detection and geographic image classification based on collection of part detectors (COPD). The COPD is composed of a set of representative and discriminative part detectors, where each part detector is a linear support vector machine (SVM) classifier used for the detection of objects or recurring spatial patterns within a certain range of orientation. Specifically, when performing multi-class geospatial object detection, we learn a set of seed-based part detectors where each part detector corresponds to a particular viewpoint of an object class, so the collection of them provides a solution for rotation-invariant detection of multi-class objects. When performing geographic image classification, we utilize a large number of pre-trained part detectors to discovery distinctive visual parts from images and use them as attributes to represent the images. Comprehensive evaluations on two remote sensing image databases and comparisons with some state-of-the-art approaches demonstrate the effectiveness and superiority of the developed framework.