Handling massive data size issue in buildings footprints extraction from high-resolution satellite images

Basit Öğe Kaydı
dc.contributor.authorAbujayyab, S.K.M.
dc.contributor.authorKaras, I.R.
dc.date.accessioned2024-09-29T16:21:20Z
dc.date.available2024-09-29T16:21:20Z
dc.date.issued2020
dc.departmentKarabük Üniversitesien_US
dc.description1st Eurasian BIM Forum, EBF 2019 -- 31 May 2019 through 31 May 2019 -- Istanbul -- 238529en_US
dc.description.abstractBuilding information modelling BIM is relying on plenty of geospatial information such as buildings footprints. Collecting and updating BIM information is a considerable challenge. Recently, buildings footprints automatically extracted from high-resolution satellite images utilizing machine learning algorithms. Constructing required training datasets for machine learning algorithms and testing data is computationally intensive. When the analysis performs in large geographic areas, researchers are struggling from out of memory problems. The requirement of developing improved, fit memory computation methods for accomplishing this computation is urgent. This paper targeting to handling massive data size issue in buildings footprints extraction from high-resolution satellite images. This article established a method to process the spatial raster data based on the chunks computing. Chunk-based decomposition decomposes raster array into several tiny cubes. Cubes supposed to be small enough to fit into available memory and prevent memory overflow. The algorithm of the method developed using Python programming language. Spatial data and developed tool were prepared and processed in ArcGIS software. Matlab software utilized for machine learning. Neural networks implemented for extracting the buildings’ footprints. To demonstrate the performance of our approach, high-resolution Orthoimage located in Tucson, Arizona state in American United States was utilized as a case study. Original image was taken by UltraCamEagle sensor and contained (11888 columns, 11866 rows, cell size 0.5 foot, 564,252,032 pixels in 4 bands). The case image contained (1409 columns, 1346 rows, and 7586056 pixels in 4 bands). The full image is impossible to be handled in the traditional central processing unit CPU. The image divided to 36 chunks using 1000 rows and 1000 columns. Full analysis spent 35 min using Intel Core i7 processor. The output performance accuracy of the neural network is 98.3% for testing dataset. Consequences demonstrate that the chunk computing can solve the memory overflow in personal computers during buildings footprints extraction process, especially in case of processing large files of high-resolution images. The developed method is suitable to be implemented in an affordable lightweight desktop environment. In addition, building footprints extracted effetely and memory overflow problem bypassed. Furthermore, the developed method proved the high quality extracted buildings footprints that can be integrated with BIM applications. © Springer Nature Switzerland AG 2020.en_US
dc.description.sponsorshipTUBITAK; Türkiye Bilimsel ve Teknolojik Araştirma Kurumu, TÜBITAKen_US
dc.identifier.doi10.1007/978-3-030-42852-5_16
dc.identifier.endpage210en_US
dc.identifier.isbn978-303042851-8
dc.identifier.issn1865-0929
dc.identifier.scopus2-s2.0-85082475148en_US
dc.identifier.scopusqualityQ4en_US
dc.identifier.startpage195en_US
dc.identifier.urihttps://doi.org/10.1007/978-3-030-42852-5_16
dc.identifier.urihttps://hdl.handle.net/20.500.14619/9683
dc.identifier.volume1188 CCISen_US
dc.indekslendigikaynakScopusen_US
dc.language.isoenen_US
dc.publisherSpringeren_US
dc.relation.ispartofCommunications in Computer and Information Scienceen_US
dc.relation.publicationcategoryKonferans Öğesi - Uluslararası - Kurum Öğretim Elemanıen_US
dc.rightsinfo:eu-repo/semantics/closedAccessen_US
dc.subjectBuildings footprints extractionen_US
dc.subjectBuildings Information Modellingen_US
dc.subjectHigh resolution satellite imagesen_US
dc.subjectMassive data sizeen_US
dc.subjectNeural networksen_US
dc.titleHandling massive data size issue in buildings footprints extraction from high-resolution satellite imagesen_US
dc.typeConference Objecten_US

Dosyalar

Koleksiyon

Scopus İndeksli Yayınlar Koleksiyonu