膨胀卷积与金字塔表达的深度学习模型用于农村建筑物提取

由于农村建筑物结构多样、空间分布复杂等特征，自动提取面临较多困难。针对该问题，本文提出采用膨胀卷积和金字塔池化表达的神经网络模型用于遥感影像中农村建筑物自动提取。在膨胀卷积神经网络模块中，通过改变孔尺寸的大小，获取不同感受野的特征信息；在金字塔表达方面，每个模块输入不同尺度的信息，且同时下采样的倍率也不同，获取多维的金字塔尺度特征；最终将提取的浅层及深层尺度特征信息进行融合，构建一个改进的适用于农村建筑物目标自动提取的深度学习模型。试验结果表明，与FCN-8s和DeepLab模型提取的结果相比，本文方法在农村建筑物提取中表现较好的性能，提取精度明显提高，且更好保留了目标边界细节信息，减少了噪声。

Rural buildings extraction based on deep learning model with dilated convolution and pyramid representation

Automatic extraction rural buildings faces many difficulties because of the diverse structure and complex spatial distribution, etc. Aiming at this problem, this paper proposes a neural network model with dilated convolution and pyramid representation for automatic extracting rural buildings in remote sensing images. In the dilated convolution neural network module, the feature information with different receptive fields is obtained by changing the size of the dilated hole. In the pyramid representation, each module inputs different scale information, and the rate of down sampling is also different, which obtains multi-dimensional pyramid scale features. Finally, the model is fused shallow feature and deep feature to construct improved deep learning model for rural building automatic extraction. Compared with FCN-8s and DeepLab models, the experiment results show that this method performs better in the extraction of rural buildings. The accuracy of the extraction is obviously improved, the boundary details are better retention and the noise is little.