利用提升树模型综合Gabor和LPQ特征进行遥感地物识别  

Boosting Tree Model with Gabor and LPQ Feature Fusion of HSI Ground Object Recognition

作　　者：姜亚楠 张春雷 张欣[3] 徐权威 张舒涛 周锐 Jiang Yanan;Zhang Chunlei;Zhang Xin;Xu Quanwei;Zhang Shutao;Zhou Rui(School of Science,China University of Geosciences(Beijing),Beijing 100083,China;Beijing Zhongdirunde Petroleum Technology Co.Ltd.,Beijing 100083,China;School of Statistics,Beijing Normal University,Beijing 100875,China)

机构地区：[1]中国地质大学(北京)数理学院,北京100083 [2]北京中地润德石油科技有限公司,北京100083 [3]北京师范大学统计学院,北京100875

出　　处：《遥感技术与应用》2022年第2期515-523,共9页Remote Sensing Technology and Application

基　　金：国家自然科学基金青年基金项目“变分法在多时滞微分方程及微分系统中的应用研究”(11601493)资助。

摘　　要：为充分融合高光谱遥感图像空间域和频率域的特征信息,提出了一种综合多尺度Gabor和LPQ特征的空谱融合遥感地物识别模型(Ms_GLPQ)。首先,在空间域上利用Gabor滤波器组,提取出遥感图像各类地物多尺度、多方向的空间邻域特征信息,以描述图像的边缘和纹理等空间结构信息;其次,在频率域上将局部相位量化(Local Phase Quantization,LPQ)算子应用于高光谱遥感图像,提取出高光谱图像的多尺度频域纹理特征,获得图像的相位不变特征描述;然后针对其中特征冗余的问题采用主成分分析(PCA)算法进行降维,再将空间域、频率域的特征进行特征融合,获得了能充分描述图像信息的特征向量;最后采用基于提升树的机器学习分类器(XGBoost、CatBoost等)进行识别。在Indian Pines、Salinas和茶树等高光谱遥感数据集上进行学习与分类测试,准确率分别为85.88%、94.42%和92.61%。实验结果表明:与传统方法相比,Ms_GLPQ模型能够提取小比例样本图像中的有效特征,取得了区分性更强的多特征区域描述子,且在采用提升树模型进行分类时效果更优,得到了比常用分类器更高的识别精度。To fully fuse the feature information in the spatial and frequency domains of hyperspectral image(HIS),a spatial-spectrum fusion HSI ground object recognition model that integrates multiscale features of Ga⁃bor and LPQ(Ms_GLPQ)is proposed.Firstly,the Gabor filter bank is used in the spatial domain to extract the multiscale and multidirectional spatial neighborhood information of various ground objects in HSI to describe the spatial structure of its edge and texture.Secondly,the Local Phase Quantization(LPQ)operator is utilized in the frequency domain to extract the multiscale frequency domain texture features,and the phase invariant fea⁃ture description of HSI is obtained.Then the Principal Component Analysis(PCA)algorithm is used to reduce the dimensionality for the problem of feature redundancy,and the features in the spatial and frequency domains are fused to obtain the feature vector that fully describes the HSI information.Finally,the classifier based on Boosting tree(XGBoost,CatBoost,etc.)are utilized for recognition.Experiments on Indian Pines,Salinas,and tea farm datasets acquire accuracy rates of 85.88%,94.42%,and 92.61%,respectively.The experimental results show that the Ms_GLPQ model can extract effective features in HSI and obtain more discriminative multi-featured region descriptors than traditional methods,and it performs better by using boosted tree model for ground object recognition and achieves higher accuracy than other classifiers.

关 键 词：高光谱遥感 多尺度分析 GABOR滤波器组 局部相位量化 提升树模型 

分 类 号：TP79[自动化与计算机技术—检测技术与自动化装置]