DeepMAO: Deep Multi-scale Aware Overcomplete Network for Building Segmentation in Satellite Imagery

Building segmentation in large-scale aerial images is challenging, especially for small buildings in dense and cluttered urban environments. Complex building structures with highly varied geometric footprints pose an additional challenge for the building segmentation task in satellite imagery. In this work, we propose to tackle the issue of detecting and segmenting small and complex-shaped buildings in Electro-Optical (EO) and SAR satellite imagery. A novel architecture Deep Multi-scale Aware Overcomplete Network (DeepMAO), is proposed that comprises an overcomplete branch that focuses on fine structural features and an undercomplete (U-Net) branch tasked to focus on coarse, semantic-rich features. Additionally, a novel self-regulating augmentation strategy, "Loss-Mix," is proposed to increase pixel representation of misclassified pixels. DeepMAO is simple and efficient in accurately identifying small and geometrically complex buildings. Experimental results on SpaceNet 6 dataset, on both EO and SAR modalities, and the INRIA dataset show that DeepMAO achieves state-of-the-art building segmentation performance, including small and complex-shaped buildings with a negligible increase in the parameter count. In addition, the presence of the overcomplete branch in DeepMAO helps in handling the speckle noise present in the SAR image modality.

1. Introduction

The availability of high-resolution satellite images from different modalities, such as Electro-Optical (EO) and Synthetic Aperture Radar (SAR), has made it easy to monitor urban environments on a large scale. Building segmentation is important in monitoring changes in the urban landscape, as buildings are critical components in these regions. It aims to classify the area occupied by the building in the image by pixel-level classification. Building information is used in many applications like change monitoring, map updating, disaster response [1], population density estimation [23], humanitarian aid, and 3-D modeling [4]. In these applications, high-resolution SAR imaging is highly beneficial, as it provides consistent information over EO imaging due to all-weather operational capabilities. However, SAR sensors have certain drawbacks such as speckle noise and less semantic information, which makes interpretation challenging for computer vision systems as well as human interpreters [20]. Automated building detection still faces significant challenges due to the diversity of buildings in terms of shapes and sizes, the complex background environment, and the complexities introduced due to SAR sensors.