深度卷积神经网络的遥感影像滑坡灾害识别

landslide hazard identification using remote sensing image based on depth convolutional neural network

为精确识别地震灾害遥感影像中的滑坡体,提出了面向对象影像分类与深度卷积神经网络相结合的方法。通过计算机自动识别九寨沟无人机拍摄的滑坡体影像,构建一种改进的最优分割尺度模型多尺度分割遥感影像,利用深度卷积神经网络提取滑坡的深度特征,分类识别滑坡体特征。结果表明,该方法对滑坡体识别的最高精度达87.68%,Kappa系数为86.34%,明显优于基于像元与深度卷积神经网络结合或面向对象分类与SVM结合的影像分类方法。

This paper proposes a method designed for accurate identification of the landslides in the remote sensing image of the earthquake disaster,a method building on the combination of object-oriented image classification and depth convolution neural network.The study involves using computer which manages to automatically identify the landslide images produced by Jiuzhaigou UAV;developing an improved optimal segmentation scale model for multi-scale segmentation of remote sensing images;extracting the depth features of landslides using depth convolution neural network and thereby classifying and identifying the landslide features.The results show that the proposed method with the landslide identification accuracy of up to 87.68%,and the Kappa coefficient of 86.34%,boasts a significant advantage over the image classification method based on the combination of pixel and depth convolution neural network or the combination of object-oriented classification with SVM.