| 摘要: |
| 为了评价主成分分析方法和决策级数据融合方法对高光谱图像地物分类结果的影响, 作者利用各分类器和主成分分量的优势, 提出了两组地物分类结果的决策级融合策略, 利用8 种常用监督分类方法对高光谱原始图像和PCA 变换后不同分量组合图分类得到的结果进行决策级融合, 并应用覆盖黄河入海口新老河道交界处的CHRIS/PROBA 高光谱图像开展实验研究。结果表明: 直接采用每类地物分类精度最高的、空穴和缝隙采用总体分类精度最高的融合策略, 在综合考虑生产者精度和用户精度的情形下, 仅使用最大似然法、支持向量机和人工神经网络3 种分类方法, 按照分类精度从高到低的顺序进行的融合分类效果最好, 总体分类精度为87.82%。与8 种监督分类方法中效果最好的最大似然法相比, 精度提高了2.7 个百分点, 同时明显减少了错分现象, 尤其是对于分布面积较小的翅碱蓬和柽柳, 滩涂被误分为翅碱蓬、芦苇被误分为柽柳的现象大大降低。 |
| 关键词: 主成分分析 高光谱图像 图像分类 决策级融合 |
| DOI：10.11759/hykx20141011012 |
| 分类号: |
| 基金项目:国家自然科学基金(41206172); 中欧国际合作龙计划项目(ID: 10470) |
|
| Research on the classification method of the hyper-spectral image based on principal component analysis and decisionlevel fusion |
|
|
| Abstract: |
| In order to evaluate the effects of PCA and decision-level fusion on the classification results of the hyper-spectral image, Herein, we took full advantage of each classifier and principal components, and put forward two decision-level fusion strategies which fused the classification results of hyper-spectral original image and different components combination after PCA transforms using eight kinds of usual supervised classification methods, and the experiment was made using the CHRIS/PROBA hyper-spectral images of the border between the new and old river courses of the Yellow River estuary. The results showed the decision-level fusion strategy that the highest classification accuracy of each type of ground objects was adopted directly and the highest overall classification accuracy was used in the holes and gaps was the best under the following conditions: taking the producer and user accuracy into comprehensive consideration; using Maximum Likelihood, Support Vector Machine and Artificial Neural Network only; fusing from the highest classification accuracy to the lowest, the overall accuracy is 87.82%. Compared with Maximum Likelihood that is the best in eight kinds of supervised classification methods, the accuracy increased 2.7%. At the same time, the incorrect classification decreased obviously, especially for the Suaeda heteroptera and Chinese tamarisk, which have a smaller area. The two situations that tidal flat was wrongly classified into Suaeda heteroptera and bulrush was misclassified as Chinese tamarisk were reduced greatly. |
| Key words: Principal Components Analysis (PCA) hyper-spectral image image classification decision-level fusion |
