摘要
从泥石流沟谷地貌条件出发,借助数字高程模型(DEM)图,对泥石流沟谷发生泥石流的概率进行预测。首先将泥石流沟谷的DEM图进行分类,分为发生过泥石流与未发生过两种;接着使用VGG与AlexNet及其对应的残差这四种神经网络对上述两种样本进行训练,实现4分类预测;最后通过结果对比,VGG能达到平均73.87%的预测正确率,其残差模型能够达到平均74.88%的预测正确率,而AlexNet与其残差的平均预测正确率仅有68%左右,实验结果表明VGG与其残差的整体性能是优于AlexNet与其残差的性能。
This paper mainly starts from the geomorphological conditions of the mudslide valley and uses the Digital Elevation Model(DEM) diagram to predict the probability of landslide in the mudslide valley. First of all, the DEM diagram of the mudslide valley is classified into two kinds of landslides and those that have not occurred, and then the four neural networks of VGG and AlexNet and their corresponding residuals are trained to achieve 4 classification predictions, and finally through the comparison of results,VGG can achieve an average prediction accuracy of 73.87%, and its residual model can achieve an average prediction accuracy of 74.88%, while the average prediction accuracy of AlexNet and its residuals is only about 68%. The overall performance of VGG and its residuals is superior to AlexNet and its residuals.
作者
袁若浩
王保云
YUAN Ruo-hao;WANG Bao-yun(Yunnan South Normal University,Kunming 650500,Yunnan;Key Laboratory of Modeling and Application of Complex Systems in Universities of Yunnan Province,Kunming 650500,Yunnan)
出处
《电脑与电信》
2022年第6期5-9,共5页
Computer & Telecommunication
基金
国家自然科学基金“基于深度迁移学习的遥感影像中泥石流孕灾沟谷识别——以云南省为例”,项目编号:61966040。
