A Road Extraction Method Based on Multi-branch Pyramid Neural Network for Remote Sensing Image

Abstract

The invention discloses a method for extracting complete road information on remote sensing images by using a convolutional neural network. A multi-branch pyramidal neural network is constructed. Through two parallel feature recovery structures and post-processing technology based on geometric structure analysis and feature point extraction, the low-level position information and high-level semantic information are fully mined to solve the problem of road extraction on satellite images. The problem of road sparsity and road breakage caused by vegetation and buildings are faced. This method comprehensively considers the sparsity of road distribution and the imbalance of samples through sparsity testing and designing a new loss function, so that the network pays more attention to sparse and difficult-to-classify road pixels. At the same time, in view of the situation where the road is covered by vegetation and buildings, this method performs automatic fracture detection and fracture connection based on geometric feature point analysis, which improves the accuracy of road extraction, and has an important role in computer vision and remote sensing road extraction scenarios. practical application value.

Description

technical field [0001] The method of the invention belongs to the field of geographic information system and computer vision, and particularly relates to extracting road information on remote sensing images by using a convolutional neural network. Background technique [0002] As an important basic geographic information data, roads play an important role in urban planning, transportation and logistics, emergency and disaster relief, and travel navigation. At present, the extraction and update of road information mainly include the following: (1) traditional surveying and mapping methods: relying on manual field measurement and recording; (2) road extraction based on GPS trajectory: collecting vehicles, taxis and People's travel trajectories to analyze road information. The above two methods require a lot of manpower and material resources, take a long time to collect data in the early stage, and have a long period of extraction and update of the entire road information, wh...

AI Technical Summary

Problems solved by technology

The technical solution adopted by the present invention is to use deep learning technology to construct a multi-branch pyramid neural network, and to solve the problem of road extraction on satellite images through two parallel feature recovery structures and post-processing technology based on geometric structure analysis and feature point extraction. The problem of road sparsity and road breakage caused by vegetation and buildings

Images