摘要
以杭州市主城区为试验区,针对建设用地与裸地空间纹理的复杂度和水体与阴影高程差异,拟采用半方差函数与Z检验结合选出的图像纹理结合高程信息等分量实现神经网络分类.结果表明,与单纯使用光谱信息相比,图像纹理的引入使总体分类精度提高约4%,加入高程信息则可以使总体分类精度提高约10%,达到82.75%,表明该方法可以应用于新数据的分类并得到相对满意的结果.
Land use of Hang Zhou city was classified from ZY-1 02C imagery in an neutral network approach using spectral,texture,and nDSM（Normalized Digital Surface Model） features.Texture features are selected from the combined use of semi-variance function and Z test.Construction and bare land were separated according to texture complexity distinction.Shadow and water were identified with the support of nDSM.Accuracy assessment indicate that addition of image textures can improve overall classification accuracy by 4% in comparison with classification using original bands solely.Furthermore,inclusion of elevation data can increase overall accuracy by 10% to 82.78%,which demonstrates the effectiveness of proposed method in the classification of 02C data.Classification result is acceptable.
出处
《浙江大学学报（工学版）》
EI
CAS
CSCD
北大核心
2013年第8期1508-1516,共9页
Journal of Zhejiang University：Engineering Science
基金
国家自然科学基金资助项目(31172023)
国家高技术产业化应用专项资助项目(2009214)
浙江省重点科技创新团队资助项目(2010R50030)
