Generalisation and Multiple Representation

Generalisation quality and conflict detection

5704.2 - Features extraction of buildings and generalization using Deep Learning

Wednesday, July 5
4:30 PM - 4:50 PM
Location: Virginia B

Map generalization is an important method to produce new maps over different scales, and automated generalization based on digital maps is still not been well solved as a world-wide difficult problem. Particularly the generalization of buildings is the most difficult task in academic field. Though the shapes of buildings generally are discrete polygons which are random and hard to describe, the shapes of man-made building are relatively regular and can be simplified as standard geometry templates. For the typical template characteristics of building representation, this study designed a series of geometry templates to abstract the building shapes according to the principles including geographic landform and human cognition, and then trained a Deep Learning network which consists of Convolutional Neural Networks and Auto Encoder extracting the features of building shapes and encoding the shapes to the combinations of features, and finally calculated the similarities of the targets and templates by Arc-cosine Similarity algorithm. Cognizing the shapes as grid image like the way human seeing the world and quantifying the cognitive way using artificial neural networks, this method conducted the building simplification by searching and matching the most similar template to replace the target building. The result showed that the test images can well be matched and generalized with the templates and the architecture of the artificial neural network designed in this paper has large potential values in the application of building generalization.

Lei Ma

Office Assistant/Teaching Assistant
Lanzhou Jiaotong University

Ma Lei, master, Lanzhou Jiaotong University. He was born in Lanzhou, China in 1989. He received the B.Eng. degree in traffic and transportation from Lanzhou Jiaotong University, China, and is studying the master's degree in Geographic Information System (GIS) in Lanzhou Jiaotong University. His research interests are Automated map generalization, Machine Learning (especially Deep Learning) and Data Science. He was responsible for the software development tasks in advisor’s programs and is good at C++, Python and web programming.

Presentation(s):

Send Email for Lei Ma

Timofey Samsonov

Leading Researcher
Lomonosov Moscow State University, Faculty of Geography, Department of Cartography and Geoinformatics

Timofey Samsonov is a leading researcher at the Department of Cartography and Geoinformatics, Faculty of Geography, Lomonosov Moscow State University, Moscow, Russia. Timofey holds a PhD in Cartography (2010) from Lomonosov MSU. His interests include generalization of spatial data, multiscale mapping, terrain mapping and analysis, spatial analysis and automation in cartography. Timofey is active in two ICA Commissions: ICA Commission on Generalisation and Multiple Representation and ICA Commission on Mountain Cartography.

Presentation(s):

Send Email for Timofey Samsonov


Assets

5704.2 - Features extraction of buildings and generalization using Deep Learning



Attendees who have favorited this

Please enter your access key

The asset you are trying to access is locked. Please enter your access key to unlock.

Send Email for Features extraction of buildings and generalization using Deep Learning