Remote sensing image building extraction method based on attention mechanism and boundary loss

Abstract

The invention relates to a remote sensing image building extraction method based on an attention mechanism and boundary loss. The method comprises the following steps: acquiring a to-be-extracted optical remote sensing image; calling the trained building extraction model; and inputting the optical remote sensing image into the building extraction model for building extraction to obtain a building extraction result of the optical remote sensing image. The encoder of the building extraction model comprises a feature extraction layer used for carrying out feature information extraction on an optical remote sensing image; the global context attention module is used for performing global context feature fusion processing on the feature information output by the feature extraction layer; and the cavity space pyramid pooling module is used for carrying out cross-scale fusion on the feature maps output by the feature extraction layer. A decoder of the building extraction model is used for carrying out resolution recovery on a feature map output by an encoder, and a loss function is a combined loss function of a boundary loss function and a Dice loss function. And the separability of the extracted building is greatly improved.

Description

technical field [0001] The present application relates to the technical field of automatic extraction of optical remote sensing image information, in particular to a method for extracting buildings from remote sensing images based on attention mechanism and boundary loss. Background technique [0002] Building extraction from optical remote sensing images is a research hotspot in the field of computer vision and remote sensing image processing in recent years. With the improvement of computer processing speed and extraction accuracy, it has been applied in urban planning, engineering drawing, urban construction and monitoring and other fields. Building extraction can be regarded as a classification problem. Each pixel in optical remote sensing images is divided into a specific category (building / non-building), which can be extended to object-oriented classification or pixel-oriented segmentation tasks. [0003] Traditional building extraction methods, such as segmentation-ba...

AI Technical Summary

Problems solved by technology

However, in the process of realizing the present invention, the inventors found that the aforementioned traditional building extraction method has the technical problem of low separability of the extracted buildings

Images