Impervious surface extraction method for high-resolution remote sensing images based on cross-sensor migration

Abstract

The invention relates to a high-resolution remote sensing image impervious surface extraction method based on cross-sensor migration. The invention combines the deep learning theory, designs a normalization layer across sensors, uses a data-driven method to hierarchically and automatically learns the normalization parameters of the radiation characteristics of the sensor itself, and eliminates the difference in characteristics between different remote sensing sensors; and designs sensor-related confrontation optimization training for efficient optimization. The semantic-related convolution parameters of the ground objects and the normalization parameters related to each sensor enhance the migration ability of the deep learning model between images from different data sources. The invention can solve the problem that the existing deep migration learning method cannot take into account the difficulty of cross-sensor migration caused by the difference of sensor characteristics, realize the migration of the deep impervious surface extraction model from high-resolution airborne images to spaceborne images, and effectively improve the Accuracy of classification and mapping across sensors transfer.

Description

technical field [0001] The invention is based on the technical processing field of remote sensing images, and particularly relates to a method for extracting impervious surfaces of high-resolution remote sensing images based on cross-sensor migration. Background technique [0002] With the rapid development of earth observation technology, massive multi-source high-resolution remote sensing images are continuously acquired by various sensors. High-resolution imagery makes smaller terrestrial objects visible and can be used to explore finer structures inside cities. Urban impervious surface refers to the type of man-made surface cover such as roads, buildings and parking lots that prevent water from penetrating into the soil. Efficient and accurate mapping and monitoring of urban impervious surfaces based on multi-source high-resolution remote sensing images can provide important information for urban planning and disaster prevention in the process of urbanization. [0003]...

AI Technical Summary

Problems solved by technology

However, the existing deep learning methods have not considered the characteristics of different remote sensing sensors, resulting in poor method transferability between different sensors

Network models trained on small-scale aerospace image impermeable surface datasets are often not suitable for large-scale aerial image classification and mapping

Images