Wetland information extraction based on UAV multispectral and oblique images

  • PDF / 5,180,286 Bytes
  • 11 Pages / 595.276 x 790.866 pts Page_size
  • 101 Downloads / 192 Views

DOWNLOAD

REPORT


GMGDA 2019

Wetland information extraction based on UAV multispectral and oblique images Yixin Du 1,2 & Yang Bai 1,2 & Luhe Wan 1,2 Received: 18 August 2020 / Accepted: 1 November 2020 # Saudi Society for Geosciences 2020

Abstract Wetland monitoring is of great significance to wetland protection. In this paper, multiscale segmentation object-oriented method is used to extract information from multispectral data and tilt image data acquired by UAV. Then, a method of multicondition difference merging based on super-pixel segmentation is proposed to extract information at a single level. The experimental results show that the overall accuracy of multilevel information extraction after multiscale segmentation is 88.03%, kappa coefficient is 86.12%, while the overall accuracy of single-level information extraction is 86.32%, and kappa coefficient is 84.12%, which shows that the improved single-level method can also achieve the accuracy of multilevel information extraction. It can solve the disadvantages of multiscale segmentation and classification time-consuming and complex inheritance relationship. Keywords UAV data . Multiscale segmentation . SLIC super-pixel segmentation . Object-oriented classification

Introduction Climate change and human activities lead to continuous reduction in the size of the wetlands. During the past 10 years, over 339.63 ha (839.244 acres) of wetlands have been lost in China, down by 10% of the total. This has aroused increasingly concern of the government. To protect the precious wetlands, people have taken various measures, but the results are difficult to monitor as the area of wetlands is too large to go deep inside. Satellite remote sensing provides a convenient alternative to carry out artificial monitoring of the wetlands. However, this technology has limited spatial and temporal resolution (Schultehostedde et al. 2007). For example, the landsat-7 imagery, which is commonly used in wetland monitoring, has the panchromatic band that achieves 15-m This article is part of the Topical Collection on Geological Modeling and Geospatial Data Analysis * Luhe Wan [email protected] 1

College of Geographical Science, Harbin Normal University, Harbin 150025, China

2

Heilongjiang Province Key Laboratory of Geographical Environment Monitoring and Spatial Information Service in Cold Regions, Harbin Normal University, Harbin 150025, China

resolution and a merely 30-m resolution in multispectral (MS) bands. When resolution is a problem, the accuracy of wetland monitoring is thus reduced. High-resolution remote sensing imagery provided by UAV-based technology is the solution to the problem (Baker et al. 2006; Chen et al. 2007). The emergence of UAV has brought a revolutionary impact on the research of ecological environment (Coops et al. 2019). The data acquired by UAV have the characteristics of high image resolution, easy interpretation of ground objects, strong timeliness of flight, and low cost. These characteristics effectively make up for the deficiencies of space remote sensing and aerial rem