Image Acquisition and Segmentation of Historic Buildings in the City of Merced Using Automation and Convolutional Neural Networks
Alberto Valle, Anaïs Guillem, David Torres-Rouff, PhD
Convolutional Neural Networks along with automation and scraping tools can accelerate the analysis and acquisition of large image datasets. These tools fit the needs to solve the main challenge in understanding the preservation of an extensive list of historic buildings in Merced. Therefore, the Detectron2 pre-trained models and Selenium Webdriver and Jsoup libraries were used for the segmentation and acquisition of +2000 images from Google maps and real estate websites. This workflow resulted to provide great functionality in finding the changes in color, materials, and even the entire removal of a building. The final objective is to implement these findings in the Arches Heritage Platform for its curation and preservation for future studies.
Mobile device: Open PDF to view the document in a new browser window.