Deep attentive fusion network for flood detection on uni-temporal Sentinel-1 data

Floods are occurring across the globe, and due to climate change, flood events expected increase in coming years. Current situations urge more focus on efficient monitoring of floods detecting impacted areas. In this study, we propose two segmentation networks for detection uni-temporal Sentinel-1 Synthetic Aperture Radar data. The first network is “Attentive U-Net”. It takes VV, VH, ratio VV/VH as input. uses spatial channel-wise attention enhance feature maps which help learning better segmentation. U-Net” yields 67% Intersection Over Union (IoU) Sen1Floods11 dataset, 3% than benchmark IoU. second proposed a dual-stream “Fusion network”, where fuse global low-resolution elevation data permanent water masks with (VV, VH) Compared previous our fusion gave 4.5% IoU score. Quantitatively, performance improvement both methods considerable. quantitative comparison method demonstrates potential networks. results further validated by qualitative analysis, demonstrate that addition mask enhances results. Through ablation experiments analysis also effectiveness various design choices Our code available Github at https://github.com/RituYadav92/UNI_TEMP_FLOOD_DETECTION reuse.