CURVELET BASED U-NET FRAMEWORK FOR BUILDING FOOTPRINT IDENTIFICATION

Abstract. This paper proposes a multiresolution based U-net composite architecture for segmentation of remotely sensed images for building footprint identification. The features derived from curvelet decompositions at different scales are augmented to capture curvilinear discontinuities of the building footprint. This increases the contextual overview of the network as the same data on multiple scales is available for feature extraction and learning. This work further analyses the effects of different multiresolution methods on wavelets and curvelets for decomposition on segmentation performance. The performance is evaluated in terms of precision, recall, F-score, mean intersection over union, overall accuracy, local and global consistency errors. It is found that the proposed method has better class-discriminating power as compared to existing methods and has an overall classification accuracy of 92.4–95.22%. On comparison with the U-Net model performance, it is observed that the proposed network can identify the building areas with higher accuracy and mean intersection over union, the best performance being with curvelet basis of multiresolution analysis.