SSDBN: A Single-Side Dual-Branch Network with Encoder–Decoder for Building Extraction

In the field of building detection research, an accurate, state-of-the-art semantic segmentation model must be constructed to classify each pixel image, which has important reference value for statistical work a area. Recent research efforts have been devoted using deep learning approaches, can further divided into two aspects. this paper, we propose single-side dual-branch network (SSDBN) based on encoder–decoder structure, where improved Res2Net is used at encoder stage extract basic feature information prepared images while module deployed decoder stage. An intermediate framework was designed new fusion methods capture more in small The decoding contains deconvolution branch and enhancement branch, are responsible capturing multi-scale enhancing high-level details, respectively. All experiments were conducted Massachusetts Buildings Dataset WHU Satellite I (global cities). proposed showed better performance than other recent achieving F1-score 87.69% IoU 75.83% with low size volume (5.11 M), internal parameters (19.8 MB), GFLOPs (22.54), Dataset.