Abstract: Abstract In this paper, we propose a road segmentation model using deep learning technology. The model is essentially a U-net with pre-trained DenseNet-169 encoder. Multi-scale and high-level semantic information are extracted effectively by dense residual learning and attention mechanism. In addition, an improved focal loss function is proposed to handle extremely imbalanced road samples. Experimental results demonstrate that our proposed road segmentation model can accurately extract complex road areas in remote sensing images and has IoU of ...
(read more)
Topics: 
Artificial intelligence
Computer vision
Remote sensing