CNN (Convolutional Neural Networks) has been proved to be an effective method for road extraction in remote sensing fields recently. D-LinkNet based on LinkNet adopted consecutive dilation convolution with different expanding rate to enlarge the receptive field without reducing the resolution of the feature-maps so that had an outstanding performance in high resolution satellite imagery road extraction. However, too many parameters make some inadequacies for D-LinkNet, which introduced LinkNet as its backbone with ResNet construction. Focused on this problem, this paper proposed a new network to improve D-LinkNet: (1) Rebuild D-LinkNet by applying DenseNet as its backbone instead of ResNet; (2) Replacing initial block with stem block in the beginning of the network. The accuracy of road extraction results from the new network which was trained with our own training dataset after data augmentation was evaluated with IoU scores. The experimental results show that the proposed network has a higher IoU scores than D-LinkNet with less parameters.