Automatic registration of urban high-resolution remote sensing images based on characteristic spatial objects

Area of Science:

• Remote Sensing
• Computer Vision
• Geospatial Information Science

Background:

• Automatic registration of high-resolution remote sensing images (HRRSIs) is challenging due to local deformations from varying angles and illumination.
• Existing methods often struggle with accuracy and efficiency in complex scenarios.

Purpose of the Study:

• To propose a novel method for automatic HRRSI registration using characteristic spatial objects (CSOs).
• To enhance the accuracy and efficiency of HRRSI registration by addressing deformation challenges.

Main Methods:

• Utilized Mask R-CNN for automatic extraction of CSOs and their positioning points.
• Developed an encoding method based on object category, relative distance, and direction for nearest neighbors.
• Applied a code matching algorithm followed by position filtering to establish control points.

Main Results:

• Achieved an 88.6% registration success rate with a maximum average error of 15 pixels.
• Demonstrated a 28.6% improvement in success rate compared to conventional local feature point methods.
• The method proves effective in handling local deformations and illumination variations.

Conclusions:

• The proposed CSO extraction and matching method offers a significant advancement in automatic HRRSI registration.
• This approach provides a more accurate and efficient solution for processing high-resolution remote sensing imagery.
• The findings contribute to improved geospatial data processing and analysis.