MultiScale Probability Map guided Index Pooling with Attention-based learning for Road and Building Segmentation

• DOI: 10.48550/arxiv.2302.09411
• Publication Date: 2023-01-01
• Authors: Shirsha Bose; Ritesh Sur Chowdhury; Debabrata Pal; Shivashish Bose; Biplab Banerjee; Subhasis Chaudhuri
• Keywords: Pooling; Benchmark (surveying)

Efficient road and building footprint extraction from satellite images are predominant in many remote sensing applications. However, precise segmentation map is quite challenging due to the diverse structures camouflaged by trees, similar spectral responses between roads buildings, occlusions heterogeneous traffic over roads. Existing convolutional neural network (CNN)-based methods focus on either enriched spatial semantics learning for or fine-grained topology extraction. The profound semantic information loss traditional pooling mechanisms CNN generates fragmented disconnected maps poorly segmented boundaries densely spaced small buildings complex surroundings. In this paper, we propose a novel attention-aware framework, Multi-Scale Supervised Dilated Multiple-Path Attention Network (MSSDMPA-Net), equipped with two new modules Dynamic Map Guided Index Pooling (DAMIP) Spatial Channel (DAMSCA) precisely extract footprints remotely sensed images. DAMIP mines salient features employing index mechanism retain important geometric information. On other hand, DAMSCA simultaneously extracts multi-scale features. Besides, using dilated convolution deep supervision optimizing MSSDMPA-Net helps achieve stellar performance. Experimental results multiple benchmark datasets, ensures as state-of-the-art (SOTA) method