BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//pretalx//talks.osgeo.org//BTZGNQ
BEGIN:VTIMEZONE
TZID:CET
BEGIN:STANDARD
DTSTART:20001029T040000
RRULE:FREQ=YEARLY;BYDAY=-1SU;BYMONTH=10
TZNAME:CET
TZOFFSETFROM:+0200
TZOFFSETTO:+0100
END:STANDARD
BEGIN:DAYLIGHT
DTSTART:20000326T030000
RRULE:FREQ=YEARLY;BYDAY=-1SU;BYMONTH=3
TZNAME:CEST
TZOFFSETFROM:+0100
TZOFFSETTO:+0200
END:DAYLIGHT
END:VTIMEZONE
BEGIN:VEVENT
UID:pretalx-foss4g-2023-academic-track-BTZGNQ@talks.osgeo.org
DTSTART;TZID=CET:20230628T110000
DTEND;TZID=CET:20230628T113000
DESCRIPTION:With increasing digitalization and automation\, there is a need
  to develop automatic methods to maintain and update public information st
 ored in spatial databases. The building register stores public\, building-
 related information and is the fundamental record for storing information 
 and other relevant data necessary for taxation\, public planning\, and eme
 rgency services about buildings. Up-to-date building footprint maps are es
 sential for many geospatial applications\, including disaster management\,
  population estimation\, monitoring of urban areas\, updating the cadaster
 \, 3D city modeling\, and detecting illegal construction cases (Bakirman\,
  et al.\, 2022.). There are many approaches for building extraction from v
 arious data sources\, including satellite\, aerial\, or drone images and 3
 D point clouds. However\, there is still a demand for developing methodolo
 gies that can extract segment\, regularize and vectorize building footprin
 ts using deep learning in and end-to-end workflow.\n\nToday\, automatic an
 d semi-automatic methods have achieved state-of-the-art results in buildin
 g footprint extraction by combining computer vision and deep learning tech
 niques. Semantic segmentation is a method for classifying each pixel in an
  image and extract building footprints from remote sensing data. In the ca
 se of building segmentation\, the goal is to classify each pixel on an ima
 ge belonging to its corresponding class. Recent advances in deep learning 
 for building segmentation have drastically improved the accuracy of the se
 gmented building masks using Convolutional Neural Networks (CNNs).\n\nRece
 ntly proposed semantic segmentation architectures include the application 
 of advanced vision transformers for semantic segmentation. GeoSeg is one o
 f the open-source semantic segmentation toolboxes for various image segmen
 tation tasks. The repository has 7 different models\, that can be used for
  either multi-class or binary semantic segmentation tasks\, including four
  vision transformers: U-NetFormer\, FT-U-NetFormer\, DCSwin\, BANet and th
 ree regular CNN models: MANet\, ABCNet\, A2FPN.\n\nThese specific methods 
 for building segmentation involve training the neural network on a labeled
  image dataset\, referred to as supervised learning. Semantic segmentation
  aims to distinguish between semantic classes in an image but does not ind
 ividually label each instance. On the other hand\, instance segmentation a
 ims at distinguishing between semantic classes and the individual instance
 s of each class. Many popular instance segmentation architectures exist\, 
 such as Mask R-CNN and its predecessors\, R-CNN\, Fast R-CNN\, and Faster 
 R-CNN. While the implementation of instance segmentation can be more chall
 enging\, the approach can be more effective in densely populated urban are
 as\, where buildings may be close or overlapping.\n\nA common problem with
  these methods is the irregular shape of the predicted segmentation mask. 
 Additionally\, the data contains various types of noise\, such as reflecti
 ons\, shadows\, and varying perspectives\, making the irregularities more 
 prominent. Further post-processing steps are necessary to use the results 
 in many cartographic and other engineering applications (Zorzi et al.\, 20
 21).\n\nThe solution for the irregularity of the building footprints is to
  use regularization. Regularization is a technique in machine learning tha
 t applies constraints to the model and the loss function during the traini
 ng process to achieve a desired behaviour (Tang et al.\, 2018). Applying r
 egularization constrains the segmentation map to be smoother\, with clearl
 y defined and straight edges for buildings. As a result\, the building foo
 tprint becomes less irregular when occluded and visually more appealing. M
 ost studies apply regularization after image segmentation. According to ou
 r knowledge\, there need to be more studies that apply regularization dire
 ctly during model training. Another alternative would be to provide an end
 -to-end workflow for regularized building footprint extraction consisting 
 of three parts: (1) segmentation\, (2) regularization and (3) vectorizatio
 n.\n\nWe propose an end-to-end workflow for building segmentation\, regula
 rization and vectorization using four different convolutional neural netwo
 rk architectures for binary semantic segmentation task: (1) U-Net\, (2) U-
 Net-Former\, (3) FT-UNet-Former and (4) DCSwin. We further improve the bui
 lding footprints by applying the projectRegularization method proposed by 
 (Li et al.\, 2021). The technique uses a boundary regularization network f
 or building footprint extraction in satellite images combining semantic se
 gmentation and boundary regularization with an end-to-end generative adver
 sarial network (GAN). Our approach will perform semantic segmentation with
  our trained models and then perform boundary regularization on the segmen
 tation masks. We aim to prove the scalability of projectRegularization on 
 a different segmentation task\, including aerial images as the data source
 . The last step in our approach is to develop a methodology for efficient 
 vectorization of the segmented building mask using open-source software so
 lutions. We aim to make the results practically applicable in any GIS envi
 ronment. The dataset used for testing our developed method will be the Map
 AI dataset used for the MapAI: Precision in Building Segmentation competit
 ion (Jyhne et al.\, 2022) arranged with the Norwegian Artificial Intellige
 nce Research Consortium in collaboration with the Centre for Artificial In
 telligence Research at the University of Agder (CAIR)\, the Norwegian Mapp
 ing Authority\, AI:Hub\, Norkart\, and The Danish Agency for Data Supply a
 nd Infrastructure.\n\nWe aim to produce better representations of building
  footprints with more regular building boundaries. After successful applic
 ation\, our method generates regularized building footprints\, that are us
 eful in many cartographic and engineering applications. Furthermore our re
 gularization and vectorization workflow is further developed into a workin
 g QGIS-plugin that can be used to extent the functionality of QGIS. Our en
 d-to-end workflow aims to advance the current research in convolutional ne
 ural networks and their application for automatic building footprint extra
 ction and\, as a result\, further enhance the state of open-source GIS sof
 tware.
DTSTAMP:20260517T153331Z
LOCATION:UBT E / N209 - Floor 3
SUMMARY:An end-to-end deep learning framework for building boundary regular
 ization and vectorization of building footprints - Simon Šanca
URL:https://talks.osgeo.org/foss4g-2023-academic-track/talk/BTZGNQ/
END:VEVENT
END:VCALENDAR