杨慧芳^1,2, 马毅^1,2, 刘荣杰², 李晓敏²
1.山东科技大学, 山东 青岛 266590;2.自然资源部第一海洋研究所, 山东 青岛 266061

摘要:
本文应用辽东湾鲅鱼圈研究区赤潮高光谱遥感数据，针对互信息RX（Reed-Xiaoli）和传统相关系数RX等单源算法赤潮检测结果不稳定，可能出现漏检和误检的问题，提出基于互信息和相关系数的决策树RX赤潮高精度检测模型，结果表明：与单源算法相比，决策树RX算法赤潮高光谱遥感检测得到的ROC曲线较稳定；与经典的非监督和监督遥感分类算法相比，决策树RX算法的错分情况明显减少，检测精度较高，总体精度达96%以上；子空间划分得到的特征波段组合，能够在不降低赤潮检测精度的前提下，检测速度提高3倍；耀斑抑制后数据1的特征波段组合赤潮检测总体精度提高了1.82%。
关键词:赤潮高光谱决策树RX算法子空间划分
DOI：10.11759/hykx20181205002
分类号:TS254
基金项目:国家自然科学基金重大项目课题“典型海洋目标多维高分辨光学遥感识别反演方法与应用验证”（61890964）；国家自然科学基金青年基金（41506204）；国家自然科学基金青年基金（61601133）

Decision tree RX hyperspectral remote sensing algorithm for high-precision and fast detection of red tide
YANG Hui-fang^1,2, MA Yi^1,2, LIU Rong-jie², LI Xiao-min²
1.Shandong University of Science and Technology, Qingdao 266590, China;2.First Institute of Oceanography, Ministry of Natural Resources, Qingdao 266061, China
Abstract:
In this paper, we apply the red tide hyperspectral remote sensing data of the Bayuquan Bay Research Area in Liaodong Bay. Because there is an instability in the red tide detection results of single-source algorithms such as mutual information Reed-Xiaoli (RX) and traditional correlation coefficient RX, and with the possible occurrence of problems such as missed and false detections, this study proposes a decision tree RX high-precision red tide detection model based on mutual information and correlation coefficient. The test results demonstrate that the ROC curve obtained by the decision fusion RX algorithm is more stable than that obtained by a single-source algorithm. As compared with the classical unsupervised and supervised remote sensing classification algorithms, a significant reduction in the misclassification of decision fusion RX algorithm is observed with higher detection accuracy. The overall accuracy of the decision fusion RX algorithm is more than 96%. The combination of feature bands obtained by the subspace partition can significantly increase the detection speed by three times without reducing the accuracy of red tide detection. The overall accuracy of the 1FBH red tide detection after the removal of the glint was 98.69%, indicating an increase of 1.82%.
Key words:red tidehyperspectraldecision fusionRX algorithmsubspace partition





PDF全文下载地址:

点我下载PDF