In the recent years, building boundary segmentation obtained significant advancement through using deep learning. The present algorithms, such as Convolutional Neural Network (CNN) are unable to detect buildings in challenging urban areas like occlusions. This study investigates the integration of U-Net and Fully Convolutional Networks (FCN) with Mask R-CNN to improve building boundary segmentation using high-resolution satellite imagery. A sequential hybrid approach has been developed for combining semantic and instant segmentation. The integration between the U-Net with Mask R-CNN has been achieved by feeding the segmentation result from the U-Net as an input into the Mask R-CNN. A similar procedure was applied in the integration of the FCN with Mask R-CNN. The integration of U-Net with Mask R-CNN resulted in an improvement in the recall by 9.9% and an increase by 4.3 % in the F1-score, demonstrating its capability in segmenting boundary precision and fine-grained details. Similarly, FCN combined with Mask R-CNN has shown an enhancement of recall by 9.9% and precision by 7.6%, assuring its capability in the capture of global context. Further analysis through comparison between integration U-Net with Mask R-CNN with results from previous studies, demonstrates that the proposed integration scheme outperforms the existing results. The performance evaluation across RGB and panchromatic datasets highlights the flexibility of these integrations by proving their efficiency in different applications. Despite the minor challenges that appeared in boundary alignment, the results brought out the potential of such hybrid models for applications in urban planning, cadastral mapping, and disaster management.