A method for extracting buildings from remote sensing images based on convolutional neural network

Abstract

The invention discloses a method for extracting buildings from remote sensing images based on a convolutional neural network, which includes three steps: (1) A feature map fusion method based on an attention mechanism is established in the building decoding stage, including multiplication and Add two ways of calculating attention weight; (2) add a branch task of building detection for joint training to improve the accuracy of the main task of building extraction; (3) add a penalty to the building edge pixels in the loss function . The present invention integrates the attention mechanism and multi-task learning into the convolutional neural network, can capture different high-level feature expressions of buildings, obtain richer feature expressions of buildings through feature fusion, and improve the accuracy of building extraction. At the same time, the present invention adds a penalty to the building edge pixels, which can effectively alleviate the problem of jagged building edges existing in the extraction results.

Description

technical field [0001] The invention belongs to the technical field of image processing, and in particular relates to a method for extracting buildings from remote sensing images based on a convolutional neural network. Background technique [0002] At present, countries around the world have launched remote sensing satellites with multiple functions to monitor the conditions of meteorology, environmental protection, agriculture, forestry, oceans, land and other resources. As one of the most important artificial targets among ground targets, buildings play an important role in urban planning, military reconnaissance, and map drawing. For example, in disaster-stricken areas, roads may be damaged in a large area. If buildings can be automatically extracted from remote sensing satellite images or aerial images, a feasible road can be quickly found, thereby speeding up rescue work. Extracting buildings from remote sensing images can be divided into the following two categories:...

AI Technical Summary

Problems solved by technology

[0005] In view of the above, the present invention provides a method for extracting buildings from remote sensing images based on convolutional neural networks. By integrating attention mechanism and multi-task learning into convolutional neural networks, different high-level feature expressions of buildings can be captured. Feature fusion can obtain richer feature expressions of buildings, improve the accuracy of building extraction, and can solve the problems of complex shape and structure, large amount of data, and difficult labeling of buildings in remote sensing images

Images