A method of target recognition from remote sensing images

  • Authors:
  • Yili Fu;Kun Xing;Xianwei Han;Shuguo Wang

  • Affiliations:
  • Sate Key Laboratory of Robotics and System, Harbin Institute of Technology, Harbin, China;Sate Key Laboratory of Robotics and System, Harbin Institute of Technology, Harbin, China;Sate Key Laboratory of Robotics and System, Harbin Institute of Technology, Harbin, China;-

  • Venue:
  • IROS'09 Proceedings of the 2009 IEEE/RSJ international conference on Intelligent robots and systems
  • Year:
  • 2009

Quantified Score

Hi-index 0.00

Visualization

Abstract

According to the characteristics of airfield and harbor from remote sensing images, a method of large target recognition based on the combination of target region and shape features is presented. First, edge detection and improved Hough transform are used to select line segments, the region including regular-array line segments in image is considered as region of interesting (ROI). ROI detection is the base for recognition. Target geometry shape is extracted from ROI using optimum threshold segmentation, which removes location effect and improves efficiency. As calculating shape principal orientations, all shapes are rotated to the same horizontally right to avoid rotation effect. The features extracted from shape implement multi-levels representation with moment features, normalized moment of inertia, length-width ratio and compact ration. Finally, feature vectors are normalized to measure similarity between target and template. Experiments show that target regions can be located accurately using ROI detection and it is effective for target recognition. Besides, the extracted features have good invariability with respect to rotation, translation and scaling, and they comprise local and overall consistency of the target, therefore, the recognition results meet expectations well.