Research on Geographic Registration of Airborne Optical Images Based on Machine Learning

This paper presents an algorithmic framework for aerial photoelectric image geo-referencing based on machine learning. The main ideas of existing multimodal image matching methods are summarized, and the advantages and disadvantages of mainstream methods in solving this problem are analyzed. Aiming at the characteristics of the 2D photoelectric image and the 3D synthetic vision registration problem, the algorithm framework is proposed by drawing on traditional methods that overcome perspective differences based on POS information and feature extraction using high-dimensional data representation. The proposed algorithm can simultaneously provide optimized POS information and geo-referenced images. The mathematical model of key steps in the calculation process is given, and the algorithm is verified using multimodal images, proving its effectiveness. The direction for future work is provided.