<p>In recent years, building detection based on images captured by remote sensing devices is one of the most important problems to be solved. Deep learning models are often used to perform these operations. Building segmentation is one of the most important resources for urban planning, transportation, geographic information systems, organization of operations after natural disasters and various military strategies. This research proposes a new semantic segmentation network, the Adaptive Building Detection Segmentation Network, which plays a crucial role in search and rescue operations after potential natural disasters. The Adaptive Building Detection Segmentation Network utilizes a sophisticated encoder-decoder redundancy block and an efficient Hybrid Attentional Atrous Convolution, both integrated into an encoder-decoder architecture. This type of architecture facilitates the effective fusion of contextual and local information, thus improving segmentation performance. The proposed model uses a complex structure that aims to derive detailed feature representations from remote sensing imagery to enable accurate building detection while minimizing the loss of contextual information. Based on this improved structure, the Adaptive Building Detection Segmentation Network retains essential details and provides higher accuracy with fewer parameters than conventional segmentation models used in other applications. Adaptive Building Detection Segmentation Network was trained and tested together with several other segmentation models on the datasets “WHU Building” and “Inria Aerial Image Labelling”. The Adaptive Building Detection Segmentation Network achieved mIoU accuracy values of 90.94% and 79.59% for these datasets. In addition, the developed Adaptive Building Detection Segmentation Network model was tested in building damage assessment studies during the Kahramanmaraş earthquake in 2023.</p>