Automated Building Extraction from an Aerial Imagery for Urban Monitoring Using a Graph-Based Deep Learning Model

Building extraction from an aerial image is essential for applications like urban development and disaster management. At present, the emergence of deep learning (DL) has greatly enhanced the effectiveness of building segmentation methods. But the segmentation and identification of images with sharp boundaries continue to face challenges due to inadequately captured high-frequency information, leading to less accurate delineation of building edges. To tackle this challenge, the present work introduces an optimization-assisted Graph convolution network with Attention guiding UNet (OGCN-AG-UNet) for extracting building information and used to identify the correlation among features and identify buildings. For suppressing the image quality defects in aerial images, between the encoder and the decoder, the GCN is integrated. The attention mechanism (AM) is used for capturing the relevant information. The metaheuristic algorithm improved crayfish optimizer (ImCO) is used for hyperparameter optimization. The evaluation is performed on two benchmark datasets, such as Massachusetts Buildings, WHU, and Inria. The suggested model achieved a better IoU of 0.962 and an accuracy of 99.12%, respectively, on the Massachusetts Buildings dataset. The experimental outcomes from building extraction datasets demonstrate that OGCN-AG-UNet enhanced the feature representation for building extraction. Furthermore, its performance in building extraction surpasses that of other existing works.