Automated Building Extraction and Fine-Grained Roofs Analysis Using Masks Former Model

Abstract

Building extraction plays an important role in many applications such as urban planning, disaster management and capital allocation layer. Traditionally, this process relies on manual identification, which is time-consuming and prone to human error. Simplifying the image by reducing resolution can improve computer performance, but causes data loss and affects capture. For small patches, segmentation is performed to merge each patch individually, and aggregated results require the use of advanced merging techniques to avoid inconsistencies. It is complex and computationally expensive. This study explores the use of Mask2Former, a deep learning model for the implementation of evacuation. Mask2Former solves the problem of segmenting high-resolution aerial imagery by leveraging the power of Transformers, known for its ability to capture far-end dependencies. Modeling multiple dependencies: The Transformer architecture is good at capturing relationships in images, allowing Mask2Former to work well with large sections and designs. Better performance with lower operating costs. Evaluation: We evaluated the performance of Mask2Former on high-resolution cloud data with pixel-level building annotations. Show the positive effects. Show that Mask2Former provides a feasible, more efficient, and useful method for extracting buildings from high resolution aerial images. understanding of adapting to real situations with different shapes and types of buildings. By leveraging the power of Transformer and evaluating its performance on a proprietary dataset, this research helps advance the field and paves the way for further research on generalizability and integration with real-world applications.