摘要: |
【目的】草原鼠害是影响草原生态平衡的重要因素,本研究基于低空遥感影像探索提取鼠害信息的最佳方案和分辨率。【方法】分别使用CART 决策树、支持向量机、最邻近、贝叶斯四种监督分类方法对高原鼠兔和高原鼢鼠两种鼠害的高分辨率无人机正射影像分类并比较其精度,再使用不同飞行高度下获取的遥感影像对鼠害信息进行信息提取。【结果】提取鼠兔鼠害信息时,基于决策树分类法的总体精度为 89.00%,kappa 系数为 0.79;支持向量机分类方法的总体分类精度为 92.00%,Kappa 系数是0.83;最邻近分类法的总体分类精度为94.00%,Kappa 系数是 0.87;基于贝叶斯分类法的混淆矩阵中得到的鼠洞的分类精度最差,鼠洞的生产者精度与用户精度都在 78.00%以下。提取鼢鼠鼠害信息时,基于决策树分类结果的总精度为 93%,Kappa 系数为 0.86;支持向量机分类结果的总精度达到 95%,Kappa 系数为 0.90;最邻近法的分类结果的总精度达到 97.00%,Kappa 系数为 0.95;Bayes分类法的总体分类精度为 98.00%,Kappa 系数达到了0.95。【结论】基于面向对象的最邻近分类法是高原鼠兔鼠害信息提取的精度最优方法,基于面向对象的 Bayes 分类法是高原鼢鼠鼠害信息提取的最佳方法。对于飞行相对高度分别为 100m、120m 和200m 三种无人机遥感影像数据,随着飞行高度的增大,影像的空间分辨率越低,其分类所需要的时间、分类精度和斑块数量均呈下降趋势。 |
关键词: 鼠害 低空遥感 遥感识别 监督分类 若尔盖草原 |
DOI: |
分类号: |
基金项目:国家自然科学基金项目“藏东南冻融水力侵蚀交错带砾石空间分布格局及对土壤侵蚀影响机制”(32060370);四川省应用基础研究项目“星机地协同的若尔盖草地鼠害遥感监测研究”(2017JY0155);四川省应用基础研究项目“基于互联网+多阶段遥感反演的区域水稻参数逐田块监测技术研究”(2017JY0284) |
|
Research on Extraction of Rodent Damage Information in Alpine Grassland Based on UAV Images ——Take Zoige as an example |
Xiongruidong1, Chengwuxue1, Xiongyudan1, Diwei1, Weijiaxuan1, Wangyongxiang1, LiuKe2
|
1.School of Geography and Resource Science, Sichuan Normal University;2.Institute of Remote Sensing Application, Sichuan Academy of Agricultural Sciences /Chengdu Branch of Remote Sensing Application Center, Ministry of Agriculture and Rural Affairs
|
Abstract: |
[Purpose] Grassland rodent damage is an important factor affecting the ecological balance of grassland. This study is based on low-altitude remote sensing images to explore the best plan and resolution for extracting rodent damage information. [Method] Use CART decision tree, support vector machine, nearest neighbor, and Bayesian four supervised classification methods to classify and compare the accuracy of the high-resolution UAV orthophotos of the plateau pika and plateau zokor. Use remote sensing images acquired at different flying altitudes to extract information on rodent damage. [Result] When extracting pika and rat damage information, the overall accuracy of the decision tree classification method is 89.00%, and the kappa coefficient is 0.79; the overall classification accuracy of the support vector machine classification method is 92.00%, and the Kappa coefficient is 0.83; the overall nearest neighbor classification method The classification accuracy is 94.00%, and the Kappa coefficient is 0.87; the classification accuracy of the mouse hole obtained from the confusion matrix based on Bayesian classification is the worst, and the accuracy of the producer and the user of the mouse hole are both below 78.00%. When extracting the information of zokor and rodent damage, the total accuracy of the classification results based on the decision tree is 93%, and the Kappa coefficient is 0.86; the total accuracy of the support vector machine classification results is 95%, and the Kappa coefficient is 0.90; the total classification results of the nearest neighbor method The accuracy is 97.00%, and the Kappa coefficient is 0.95; the overall classification accuracy of the Bayes classification method is 98.00%, and the Kappa coefficient is 0.95. [Conclusion] The object-oriented nearest neighbor classification method is the best method for the accuracy of the plateau pika and rodent damage information extraction, and the object-oriented Bayes classification method is the best method for the plateau zokor rodent damage information extraction. For three types of UAV remote sensing image data with relative flight altitudes of 100m, 120m, and 200m, as the flight altitude increases, the spatial resolution of the image decreases, and the time required for classification, classification accuracy, and number of patches are all Shows a downward trend. |
Key words: Rodent damage Low altitude remote sensing Remote sensing recognition Supervised classification Zoige Grassland |