Road extraction method and system based on high-order spatial information global automatic perception

Abstract

The invention discloses a road extraction method based on high-order spatial information global automatic perception. The method comprises the steps of generating spatial features according to presetdepth convolution layer model coding; generating a road extraction feature resource pool by adopting bilinear pooling; carrying out weighted feature space redistribution according to the road extraction feature resource pool; and enabling the weighted feature space redistribution to complementarily select features from the whole road extraction feature resource pool for redistribution according tothe requirements of each local feature position of the road and the background. The invention also discloses a road extraction system based on high-order space information global automatic perception. According to the invention, features generated by coding and global spatial information are integrally and organically combined; the method is advantaged in that problems of spatial information lossand discontinuity caused by local spatial information learning are avoided, road extraction evaluation indexes are effectively improved, model parameters are reduced, operation time is reduced, intelligent segmentation extraction of the road area is realized, and the method is also suitable for semantic segmentation of other objects.

Description

technical field [0001] The invention relates to remote sensing image processing technology, neural network algorithm, and computer vision semantic segmentation technology, in particular to a road extraction method and system based on global automatic perception of high-order spatial information. Background technique [0002] The problem domain covered in this paper is the extraction of road regions from remote sensing imagery. Road segmentation based on remote sensing images has a wide range of application scenarios in digital map generation, road network update, urban planning, automatic driving, path planning, road navigation, road damage detection, natural disasters, emergency rescue and other fields. [0003] Semantic segmentation of roads is a challenging task. Different from extracting road skeleton information, each pixel belonging to the road needs to be marked as a road, and the rest is marked as the background, which belongs to the problem of binary semantic segme...

AI Technical Summary

Problems solved by technology

First, the discontinuity of the kernel (kernel) is used in the hole convolution, resulting in not all pixels participating in the calculation, loss of information continuity and globality; second, the multi-scale feature fusion module increases the model parameters

Images