| 摘要: |
| 本文提出了一种随机森林(random forest,RF)模型和Pearson相关系数相结合的RF-Pearson模型特征优选方法。以多时相Sentinel-2影像为数据源,提取多时相多特征;利用RF-Pearson模型进行特征选择,筛选出特征重要性得分较高且相关性较小的特征作为优选特征,参与黄河三角洲湿地信息提取;最后将分类结果与多时相全特征和随机森林模型优选特征进行比较。实验表明:特征优选能够提高湿地信息的提取效果,基于RF-Pearson模型特征优选方法的分类精度最高,表明了特征优选方法的有效性以及特征优选在湿地分类方面的优势。 |
| 关键词: 黄河三角洲 多时相 特征优选 Pearson |
| DOI：10.11759/hykx20220626001 |
| 分类号:TP79 |
| 基金项目:中央高校基本科研业务费专项资金资助(22CX01004A-8);国家自然科学基金面上项目(62071492) |
|
| Wetland information extraction based on multifeature optimization of multitemporal Sentinel-2 images |
|
SHENG Hui1, WEI Jing-jing1, HU Yao-dong2, XU Ming-ming1, CUI Jian-yong1, ZHENG Hong-xia1
|
|
1.College of Oceanography and Space Information, China University of Petroleum(East China), Qingdao 266580, China;2.Qingdao Municipal Institute of Surveying and Mapping, Qingdao 266033, China
|
| Abstract: |
| This paper presents a method for feature optimization of the RF-Pearson model based on the combination of the random forest model and Pearson correlation coefficient. The multitemporal Sentinel-2 image was used as the data source for extracting multitemporal features. Furthermore, the RF-Pearson model for feature selection was employed to select the features with a high importance score and low correlation of feature variables as the preferred features, participating in the wetland information extraction in the Yellow River Delta. Finally, the classification results were compared with multitemporal full and random forest model optimization features. Experiments revealed that feature optimization can enhance the extraction effect of wetland information, and the feature optimization method based on the RF-Pearson model had the highest classification accuracy, indicating the efficacy of the feature optimization method and the benefits of feature optimization in wetland classification. |
| Key words: Yellow River Delta multitemporal feature optimization Pearson |
