摘要
针对湿地植被精细分类的研究较少、分类精度不高的问题,提出了面向对象随机森林湿地植被分类方法。面向对象分割技术可减少"椒盐效应",随机森林分类算法具有高准确度、抗噪能力强、性能稳定等优势。鉴于此,通过调整面向对象的分割参数与随机森林中树的深度、个数等,构建了最优的面向对象随机森林分类模型。另外,选择了支持向量机分类算法和决策树分类算法作对比实验。实验结果显示,面向对象随机森林分类算法的总体精度达到88.3%,明显高于支持向量机算法和决策树算法,能够有效提高湿地植被分类的精度。
There are few studies on the fine classification of wetland vegetation, and the classification accuracy is not high. To solve the problem, object-oriented random forest method is proposed in this paper. Object-oriented segmentation technology reduces salt and pepper effect. Random forest classification algorithm, as the most popular of the mining algorithm, has high accuracy. It also has the ability of noise immunity, stable performance. For that reason, the best random forest classification model is built by adjusting object-oriented segmentation parameters, the depth and the number of random forest key parameters in this paper. The SVM algorithm and decision tree classification algorithm are selected for comparison. Experimental results show that the classification accuracy of wetland vegetation by object-oriented random forest method is 88. 3%, which is significantly higher than SVM and the Decision Tree algorithm. It proves that object-oriented random forest classification can effectively improve the accuracy of wetland vegetation classification.
出处
《遥感信息》
CSCD
北大核心
2018年第1期111-116,共6页
Remote Sensing Information
基金
国家自然科学基金重点项目(41330750)
国家自然科学基金面上项目(41371406)
关键词
随机森林
湿地植被
面向对象
分类
高精度
random forest
wetland vegetation
object-oriented
classification
high accuracy
