The invention discloses a robust multi-mode remote sensing image matching method and system, and various local feature descriptors are integrated to carry out multi-mode remote sensing image automaticmatching. The method comprises firstly, carrying out dense grid sampling on images, extracting a local feature descriptor of each lattice point to form a three-dimensional dense feature expression graph, and then establishing a fast matching similarity measurement in a frequency domain by utilizing a three-dimensional Fourier transform on the basis of the feature expression graph, and finally, carrying out homonymous point identification by means of the template matching. In addition, the invention further provides a new dense feature expression method channel feature of oriented gradient (CFOG) according to the matching method, the channel feature of oriented gradient (CFOG) is superior to the existing dense feature expression modes based on HOG, LSS, SIFT, SURF in the aspects of matching performance and calculation efficiency. According to the invention, the non-linear radiation difference of multi-mode images of the visible light, the infrared ray, the laser radar, the synthetic aperture radar and the map can be effectively overcome, the homonymous point can be rapidly and accurately identified among the images, and the automatic matching of the images can be realized.