基于卷积神经网络的不良地质体识别与分类  

Identification and Classification of Adverse Geological Body Based on Convolution Neural Networks

作　　者：陈冠宇[1] 安凯[1] 李向[1] 

机构地区：[1]中国地质大学(武汉)计算机学院,武汉430074

出　　处：《地质科技情报》2016年第1期205-211,共7页Geological Science and Technology Information

基　　金：湖北省自然科学基金项目"基于高空莘分辨率RS;Li DAR和GPR的多源数据融合办法研究"(2012FFB6403)

摘　　要：西部大开发战略实施以来,西部地区,尤其是新疆等地,修建高速公路成为近年来的首要任务。但是,西部海拔较高,地理环境恶劣,像沙害、盐渍土、冻土和荒漠等特殊地质体广泛分布。以新疆尉犁县罗布人村寨为研究区域,针对当地典型的不良地质体遥感影像特征,重点探讨了深度学习算法中的卷积神经网络方法在不良地质体识别与分类中的应用,实验结果对比分析表明:与K-均值分类器、SVM分类器和贝叶斯分类器对比分类精度,当样本数量较少时卷积神经网络方法优势不明显,当训练样本足够大时,其分类精度达到90%左右,优势非常明显。Since China＇s implementation of the western development strategy, in the western region, espe- cially in Xinjiang, the construction of highways become a top priority in recent years. However, the western high ahitude,harsh geographical environment, the special geological body such as comprehensive, saline soil,frozen soil and widely distributed desert always block the development. This article, taking Yuli Rob village of Xinjiang as the study area, studies the application of the convolution neural network method in deep learning algorithms for the identification and classification of adverse geological bodies based on the remote sensing image characteristics of local typical adverse geology. The experimental results analysis showed that. compared with K-mean classifier,SVM classifier and Bayesian classifier classification, the ad- vantage of convolution neural network is not obvious with small number of samples, but its advantage is remarkable, when the training sample is large enough, with the classification accuracy reaching about 90%.

关 键 词：遥感影像 不良地质体 深度学习 分类方法 神经网络 

分 类 号：P467[天文地球—大气科学及气象学]