Open AccessChange detection of built-up areas based on ensemble learning
Author(s) -
Lei Chen,
Yingcheng Li,
Zhongyuan Geng,
Xilin Li,
Yanli Xue,
Guangliang Wang,
Yanhui Wang
Publication year - 2021
Publication title -
iop conference series. earth and environmental science
Language(s) - English
Resource type - Journals
eISSN - 1755-1307
pISSN - 1755-1315
DOI - 10.1088/1755-1315/734/1/012029
Subject(s) - computer science , change detection , reliability (semiconductor) , data mining , feature (linguistics) , artificial intelligence , field (mathematics) , ensemble learning , construct (python library) , machine learning , encoder , pattern recognition (psychology) , mathematics , power (physics) , linguistics , physics , philosophy , quantum mechanics , pure mathematics , programming language , operating system
The paper presents a new method for the change detection of built-up areas based on ensemble learning. The method selects the feature generation part of VGG network to replace the encoder of UNet network, and makes full use of the advantages of VGG network in feature interpretation to construct a new network VGG-UNet, so as to improve the information extraction accuracy of built-up areas. First, the slice sample sets containing built-up areas and non-built-up areas were established. Secondly, the model was trained and tested to obtain several optimal models with high precision. Thirdly, these optimal models were used to extract the change information of built-up areas in the experimental area. Finally, the results were ensemble optimized and the accuracy was verified by comparing with the field measurements. The experimental results in Xinjiang demonstrate the advantages, feasibility and reliability of the proposed method. Moreover, the ensemble optimization of multiple model results can further improve the change detection accuracy of built-up areas.