Multi-source remote sensing image water body detection method based on iterative evolution of multi-view depth network

Abstract

The invention discloses a multi-source remote sensing image water detection method with multi-view depth network iterative evolution. In the training stage, the present invention divides the original data set into multiple non-overlapping sub-data sets, uses different sub-data sets to train deep semantic segmentation networks of different perspectives, then uses the multi-view deep semantic segmentation network to collaboratively update labels, and uses The updated labels retrain the multi-view deep semantic segmentation network, and a good deep semantic segmentation network can be obtained after multiple iterations. In the testing phase, a multi-source remote sensing image was predicted by a multi-view deep semantic segmentation network, and then voted to produce the final water body detection result. This multi-view deep network iterative evolution of multi-source remote sensing image water detection method can effectively solve the problem of low resolution of training data water labels and high noise, which leads to the reduction of water detection accuracy based on deep learning, and improve the accuracy of water detection.

Description

Technical field [0001] The present invention belongs to the cross-section of remote sensing interpretation and artificial intelligence, and specifically, the present invention relates to a multi-source remote sensing image water detection method for multi-view depth network iterative. Background technique [0002] The water detection of remote sensing images is important in flood disaster assessment, water resources value estimation, ecological environment protection, etc. Modern society is increasingly attaching importance to the environment, and people's requirements for water information are increasingly strict. The remote sensing image is short, and the imaging accuracy is increasing, high-resolution remote sensing images have become the focus of research workers. [0003] Remote sensing imaging water detection method, mainly divided into a water body method based on image spectrum characteristics, based on a hydrodynamic detection method of a classifier, and a deep study of ...

AI Technical Summary

Problems solved by technology

[0004] The present invention mainly solves the problem that the water body label resolution of the training data is low and the noise is too much to reduce the accuracy of the water body detection based on deep learning. Under the premise of low resolution and high noise of water body labels, multi-view iterative training of deep semantic segmentation network can effectively improve the accuracy of water body detection

Images