Heterogeneous remote sensing image block matching method and system based on descriptor consistency constraint

Abstract

The invention discloses a heterogeneous remote sensing image block matching method and system based on descriptor consistency constraint, and the method comprises the steps: guiding low-level feature learning through the knowledge which is obtained through high-level feature learning and comprises more similarity sorting, and optimizing a feature network through the richer supervision information in high-level features; therefore, the precision of different-source remote sensing image matching is improved. When the difference between different-source images is eliminated, a large amount of discrimination information is easy to lose, the discrimination of different-source image features is influenced, the matching error of the different-source images is increased, and the discrimination of the different-source image features is kept by means of reconstruction constraint, so that the matching precision of the different-source remote sensing images is improved. The method comprises the following steps: making a data set; preprocessing the image; extracting image block features by using a different-source image feature learning network; using a heterogeneous shared feature learning network to extract shared features of the image blocks; reducing the dimension of the network; reconstructing the image block with a decoder; optimizing the network; predicting a matching probability; and evaluating the network performance.

Description

technical field [0001] The invention belongs to the technical field of image processing, and in particular relates to a heterogeneous remote sensing image block matching method and system based on descriptor consistency constraints. Background technique [0002] Image matching refers to predicting the relationship between two images, matching or not matching. The two images to be matched often contain the content of the same scene, but they may be acquired at different times, through different sensors, from different perspectives, and under different lighting conditions. The different imaging mechanisms of heterogeneous images lead to great differences in the representation of heterogeneous images for the same scene or the same target, which greatly increases the difficulty of heterogeneous image matching. Because heterogeneous image matching not only needs to consider illumination transformation, geometric transformation, and viewing angle change, but also needs to conside...

AI Technical Summary

Problems solved by technology

There is also an exponential triplet loss function based on HardNet, which is better than hardnet, but it does not constrain the intermediate features, and the network is completely shared, which is not completely suitable for heterogeneous image matching.

Images