融入地形信息的滑坡场景识别

Landslide Scenes Recognition Integrated with Terrain Information

当前滑坡的自动识别方法以深度学习为主,数据源通常是高分辨率遥感图像或高精度DEM数据,结合高分辨率影像和高精度DEM,利用深度学习的方法能够获取到较高的识别效果。但是针对滑坡场景而言,高分辨率的地形数据往往难以获取。本文考虑融合谷歌地球影像与低精度地形数据来实现滑坡场景的自动识别,通过设计数据源类别注意力机制模块来融合影像特征和地形特征后再进行最终的分类。对比卷积神经网络仅通过谷歌影像进行识别的方式,本文提出的融合方法在识别精度、准确率、召回率和F1值这4项指标上均有明显提高。实验结果表明,通过本文所提出的方法将谷歌影像和低精度地形特征融合后再进行滑坡场景的自动识别是可行的。

At present,the automaticrecognition methods of landslideare mainly based on deep learning,and the data source is usually high-resolution remote sensing images or high-precision DEM data.Combining high-resolution images and high-precision DEM data,the use of deep learning method can obtain high recognition results.However,for landslide scenes,high-precision terrain data are often difficult to obtain.This paper considers the fusion of Google Earth images and low-precision terrain data to realize automatic recognition of landslide scenes.Through the design of the data source category attentionmechanism module to fuse image features and terrain features,the final classification is performed.Compared with the Convolutional Neural Network method of only using Google Earth images for recognition,the fusion method proposed in this paper has a significant improvement in the four indicators of accuracy,precision,recall and F1value.The experimental results show that the method proposed in this paper is feasible to automatically recognize landslide scenes by fusing Google images with low-precision terrain features.