【摘要】 针对槟榔、椰子和胡椒等海南特色经济作物结构复杂以及碎片化分布问题，利用随机森林算法，探究基于GF-2数据的特色经济作物精细分类框架。采用GF-2融合影像，通过影像分割、特征分析与筛选处理，分别采用随机森林、支持向量机和K最近邻算法，对多对象特征融合数据进行分类，并建立混淆矩阵对结果进行精度评价：(1)当分割尺度为138、形状为0.3、紧致度为0.5时，各分割对象间能够实现较高的区分度；(2)基于多对象特征融合后的随机森林分类结果，总体精度和Kappa系数分别可达84%和0.81;(3)随机森林相较于支持向量机和K最近邻算法，总体精度分别提高了8%和12%,Kappa系数分别提升了0.1和0.12。研究结果为海南地区特色经济作物遥感精细分类与槟榔黄化病研究提供了技术支持。更多还原

【Abstract】 To address the complex structure and fragmented distribution of specialty economic crops such as betel nut, coconut, and pepper in Hainan, this study explored a detailed classification framework based on GF-2 satellite data using the Random Forest algorithm. GF-2 fused imagery underwent image segmentation, feature analysis, and feature selection. Classification was conducted using the Random Forest, Support Vector Machine, and K-Nearest Neighbor algorithms on multi-object feature-fused data. A confusion matrix was used to evaluate classification accuracy. The segmented objects exhibited high differentiation when the segmentation scale was set to 138,with a shape parameter of 0.3 and a compactness parameter of 0.5. The Random Forest classification results based on multi-object feature-fused data achieved an overall accuracy of 84% and a Kappa coefficient of 0.81. Comparing with the Support Vector Machine and K-Nearest Neighbor algorithms, the overall accuracy improved by 8% and 12%,and increased the Kappa coefficient by 0.1 and 0.12 by using Random Forest, respectively. The results provide technical support for the classification of specialty economic crops using remote sensing and contribute to the scientific management of agriculture in Hainan.更多还原