共查询到18条相似文献,搜索用时 296 毫秒
1.
高光谱遥感影像维数高、数据量大、波段之间的相关性强,分类时易出现"Hughes"现象,因此在分类过程中如何有效减小数据处理过程中的计算量,又保证原始数据重要的地物信息不丢失具有重要的意义。压缩感知理论可通过远低于耐奎斯特的采样率和少量观测数据实现信号的精确重构,具有对硬件读写要求低、图像恢复效果好等优势。通过利用基于小波变换的压缩感知算法对黄河口地区的高光谱影像进行图像重构,然后分别采用SVM算法、最大似然法以及神经网络分类法对重构后的影像进行分类,并对分类结果的精度分别从空域和小波域、不同的测量值等维度进行了分析和比较。结果表明:(1)压缩感知理论重构后的影像保留了原始影像的基本信息,保证了分类精度;(2)SVM算法的分类精度最好,空域和小波域的分类精度基本一致;(3)分类精度随测量值的增加先逐渐提高,然后趋于稳定。 相似文献
2.
提出了一种融合监督分类与非监督分类结果的高光谱遥感影像分类新方法——众数赋值分类法。采用ISODATA非监督分类方法对高光谱遥感影像进行分类,并对非监督分类结果的图斑进行标记,同时用最大似然法(ML)和支持向量机(SVM)法进行监督分类,然后以监督分类结果对非监督分类后各斑块进行类别赋值。方法是:统计每个非监督分类斑块中由监督分类所获得的各类别像元数及所占比例,将非监督分类斑块的类别赋予所占比例最高的监督分类结果的类别,最终获得高光谱图像分类结果。研究表明:(1)非监督分类类别数量大于10时,其与ML分类结果融合的总体分类精度和Kappa系数均较监督分类法的分类结果好;(2)ML和20个类别的ISODATA分类结果融合的总体精度最高,为87.35%,比单独ML的总体精度高约2个百分点;(3)SVM和10个类别的ISODATA分类结果融合的总体精度提高最大,较SVM的总体精度提高近3个百分点;(4)随着非监督分类类别数量的增多,分类结果的总体精度呈现由低到高再到低的变化过程。 相似文献
3.
4.
5.
在光谱规则分类算法(Spectral Rule-based Classifier, SRC)基础上考虑大气校正对遥感影像光谱反射率的影响,提出了一种改进光谱规则的分类算法(Modified Spectral Rule-based Classifier, MSRC),从地物光谱响应曲线和光谱指数两个方面来修正光谱规则集,通过规则细化和补充、阈值改正优化光谱类别。以珠江三角洲海岛(佳蓬、淇澳)和海岸带(荃湾、惠东)的Landsat 8影像作为实验数据,对比了大气校正前后波段反射率和地物光谱响应曲线,分析了改进后MSRC算法的地物分类结果和精度,并与原SRC算法、最小距离分类(MDC)算法、最大似然分类(MLC)算法、支持向量机分类(SVM)算法、神经网络分类(NNC)算法以及基于光谱指数的算法等多种地物分类算法进行比较。结果表明,4组实验数据的MSRC算法分类结果总体精度分别为87.66%、82.38%、77.67%和80.05%,高于SRC、MDC、MLC和基于光谱指数的分类算法,在无需人工标注训练数据集的前提下接近SVM和NNC算法的分类精度。MSRC算法适用于海岛和海岸带的Lands... 相似文献
6.
高光谱遥感图像分类精度对比研究 总被引:5,自引:0,他引:5
根据国家重点基础研究发展规划项目"干旱、半干旱地区的环境动力机制重大理论基础研究"的需要,开展各种针对高光谱图像的经典分类精度对比研究.采用ENVI软件中常规高光谱分类算法和自编的模糊算法程序,对3个典型的试验区进行对比试验.结果表明,经典分类算法中的光谱角填图(SAM)、欧氏最小距离、马氏距离分类精度较高,模糊识别法中的相关系数法、绝对指数法、相对指数法、夹角余弦法的精度均比较高而且比较稳定;其它方法精度比较低.虽然有多种因素影响分类的精度,但是只要措施适当,还是可以取得比较满意的分类精度. 相似文献
7.
对覆盖黄河口滨海湿地的PROBA CHRIS高光谱遥感影像进行包络线去除变换,采用6种常用的基于光谱特征空间的监督分类算法对变换前后的影像数据进行滨海湿地典型地物分类,通过目视对比分析和定量分析相结合的方法分析比较变换前后的分类结果,评价包络线去除方法对该类算法影响的效果和能力。结果表明,包络线去除方法能够提高部分监督分类算法针对滨海湿地典型植被类型的区分和识别能力;但由于滨海湿地内具有面积较大的裸滩和浑浊水体,这两类地物在影像中的光谱特征相近,而包络线去除方法并不能解决二者的误分问题,因此并不能提高该类算法针对CHRIS高光谱遥感影像的总体分类精度。 相似文献
8.
9.
为提高遥感影像融合质量,提升资源一号(ZY-1 02D)高光谱遥感影像滨海湿地植被分类精度,提出将ZY-1 02D高光谱影像与空间分辨率为10 m的哨兵2号(Sentinel-2)影像进行Brovey融合,并通过搭建AlexNet卷积神经网络对ZY-1 02D高光谱影像和Brovey融合影像的滨海湿地植被进行分类,与支持向量机、随机森林和BP神经网络分类算法进行精度对比。研究结果表明:经Brovey融合后,AlexNet、支持向量机、随机森林和BP神经网络算法的植被分类总体精度分别提高15.60%、7.00%、14.80%和10.00%,Kappa系数提高了21.35%、9.93%、18.97%、12.85%;基于Brovey影像融合与AlexNet算法的植被分类精度最高,总体精度为92.40%,Kappa系数为89.42%。空谱融合配合AlexNet卷积神经网络有效解决了高光谱遥感影像在滨海湿地植被分类应用中精度较低的问题,为滨海湿地植被资源动态监测提供技术和方法支撑。 相似文献
10.
联合光谱和纹理特征的滨海湿地高光谱深度学习分类—以黄河三角洲湿地为例 总被引:2,自引:0,他引:2
本文基于CHRIS高光谱遥感影像,发展了一种结合地物光谱特征和多纹理空间特征信息,采用双全链接的8层深度卷积神经网络分类算法对滨海湿地高光谱影像进行遥感地物分类,并在黄河口滨海湿地进行了应用。结果表明:1)基于测试样本数据,联合光谱特征和K-L变换的纹理特征信息,采用DCNN模型方法展现了高的分类精度,精度高达99%;2)利用光谱特征和全纹理特征的精度比仅使用光谱特征和光谱特征联合K-L变换后纹理特征的分类精度低。利用K-L变换后的光谱特征和纹理特征的DCNN分类精度达到99.38%,相比于使用全纹理特征信息的精度提高了4.15%;3)基于验证图像,发展的DCNN分类方法精度优于其他算法,DCNN方法总体分类精度为84.64%,Kappa系数为0.80;4)相比于浅层分类方法,本文发展的DCNN模型分类算法保证了所有地物类型的分类精度更加均衡,保持了主要地物类型的分类精度几乎不变,同时提高了滩涂和农田的精度。基于DCNN模型,潮滩和农田的分类精度分别达到79.26%和56.72%。比其它浅层分类方法提高了2.51%和10.6%。 相似文献
11.
12.
This article studies the effect of airborne lidar (surface) elevation data on the classification of multispectral IKONOS images over a coastal area. The lidar data and IKONOS images are treated as independent multiple bands to conduct the classification. To do so, the lidar elevation data is first resampled to the same ground spacing interval and stretched to the same radiometric range as the IKONOS images. An unsupervised classification based on the ISODATA algorithm is then used to determine a class schema of six classes: road, water, marsh, roof, tree, and sand. Training sites and checking sites are selected over the lidar-IKONOS merged data set for the subsequent supervised classification and quality evaluation. The complete confusion matrices and average quality indices are presented to assess and compare the classification results. It is shown that the inclusion of the lidar elevation data benefits the separation of classes that have similar spectral characteristics, such as roof and road, water and marsh. The overall classification errors, especially the false positive errors, are reduced by up to 50%. Moreover, by using the lidar elevation data, the classification results show more realistic and homogeneous distribution of geographic features. This property will benefit the subsequent vectorization of the classification maps and the integration of the vector data into a geographical information system. 相似文献
13.
Combining Lidar Elevation Data and IKONOS Multispectral Imagery for Coastal Classification Mapping 总被引:1,自引:0,他引:1
This article studies the effect of airborne lidar (surface) elevation data on the classification of multispectral IKONOS images over a coastal area. The lidar data and IKONOS images are treated as independent multiple bands to conduct the classification. To do so, the lidar elevation data is first resampled to the same ground spacing interval and stretched to the same radiometric range as the IKONOS images. An unsupervised classification based on the ISODATA algorithm is then used to determine a class schema of six classes: road, water, marsh, roof, tree, and sand. Training sites and checking sites are selected over the lidar-IKONOS merged data set for the subsequent supervised classification and quality evaluation. The complete confusion matrices and average quality indices are presented to assess and compare the classification results. It is shown that the inclusion of the lidar elevation data benefits the separation of classes that have similar spectral characteristics, such as roof and road, water and marsh. The overall classification errors, especially the false positive errors, are reduced by up to 50%. Moreover, by using the lidar elevation data, the classification results show more realistic and homogeneous distribution of geographic features. This property will benefit the subsequent vectorization of the classification maps and the integration of the vector data into a geographical information system. 相似文献
14.
15.
16.
17.
18.
A source that emits a constant frequency tone and moves at a constant course and speed can be localized through measurements of the Doppler shifted frequencies (DSF). With five unknowns, namely, the rest frequency and the positions and speeds in the x-y directions, five separate sensors would normally be necessary to give five DSP measurements for instantaneous localization. The equations are nonlinear, and the standard solution is by grid search or iteration. The high dimensionality leads to a large computational requirement. By incorporating DSF rates, a quantity available from frequency line trackers, a one-dimensional grid search solution is possible which requires only three sensors and reduces the computational load. The derivation of the grid search technique is given, together with simulation results. The conclusion is that at high signal-to-noise ratios (SNR), the scheme reaches the three-sensor Cramer-Rao lower bound; at lower SNR's or with increased sensors, the grid search answer is a good initializer for a nonlinear optimization algorithm that gives a maximum likelihood estimate 相似文献