High-resolution remote sensing water extraction based on improved U-net

Abstract:
Aiming at the problem that water body extraction from high-resolution remote sensing images is easily interfered by irrelevant information such as vegetation, buildings, roads, etc, this paper proposes an improved model to better improve the effect of water extraction. In this paper, we introduce an efficient channel attention mechanism in the coding region to enhance target features. Dilated convolution and pyramid pooling modules were introduced to expand the receptive field without losing the resolution, and multi-scale context information was captured by different dilation rates. In this paper, we use the self-made GF-2 remote sensing image water dataset for training and evaluation.The experimental accuracy rate reaches 98.47%, the F1-score and IoU are 0.9311 and 0.9829 respectively. Compared with the FCN model and the original Unet model, the proposed model avoids the interference of vegetation and buildings, and has the advantages of high precision and low noise.