High-resolution image building extraction precision processing method based on prior vector guidance

Abstract

The invention provides a high-resolution image building extraction precision processing method based on prior vector guidance, and the method comprises the steps: carrying out the data preprocessing of a prior vector and high-resolution remote sensing image data based on grating vector consistency analysis, and obtaining a data set after geometric registration and correction; fully considering spectral and spatial characteristics of ground features based on an FNEA image segmentation algorithm, carrying out image fine segmentation on a high-resolution remote sensing image within the constraint of a prior vector boundary by setting scale parameters to obtain a series of ground feature category image spots with different sizes, and obtaining ground feature category segmented image spots under grating vector suite; and obtaining a final result of building target extraction by adopting a majority voting method, enhancing the homogeneity and integrity of a building object roof extraction result within the constraint of a priori GIS land use vector boundary, and obtaining a deep network model building extraction precision processing result under the constraint of the priori GIS land use vector. According to the method, the building extraction quality of the high-resolution remote sensing image based on the deep learning model is remarkably improved.

Description

technical field [0001] The invention relates to the technical field of remote sensing applications, in particular to a method for extracting and fine-processing high-resolution image buildings based on priori vector guidance. Background technique [0002] In recent years, with the development of computer computing power and deep learning algorithms, building target detection and recognition and semantic segmentation based on convolutional neural networks have gradually surpassed traditional algorithms, greatly improving the accuracy of building target extraction in massive remote sensing images. Rate. At present, the semantic segmentation methods based on deep learning are divided into two categories, one is the candidate region-based learning method represented by the R-CNN (Region-based Convolutional Network) series, and the other is represented by FCN and U-Net Segmentation algorithms that can be learned end-to-end. The first type of R-CNN-based series of algorithms hav...

AI Technical Summary

Problems solved by technology

Simple splicing and fine processing cannot avoid the "banding effect" caused by edge splicing, but will lead to misuse of image edge context information and inaccurate processing results

Images