基于3-D Gabor纹理特征和GBDT分类器的高光谱分类方法

A hyperspectral classification method based on 3-D Gabor texture feature and GBDT classifier

为了有效提取高光谱图像的空间和光谱维特征,获得准确率和分类效率俱佳的方法,利用52个不同方向和频率的3-D Gabor滤波器提取图像的纹理特征,结合梯度优化决策树分类器(GBDT)完成高光谱图像分类。结果表明3-D Gabor+GBDT方法的分类准确率高于CNN算法、Gabor以及EMAP为纹理特征的方法,且高于CNN和以SVM为分类器的方法。虽然3-D Gabor+GBDT建模训练时间长,但是该方法在保持高准确率的前提下,分类效率依然较高,适合大规模高光谱图像的在线分类场景。

In order to extract spatial and spectral dimension features of hyperspectral images effectively,and obtain a method with good accuracy and classification efficiency,523-D Gabor filters with different directions and frequencies are used to extract texture features of images,and the gradient optimization decision tree classifier(GBDT)is used to complete the classification of hyperspectral images.The results show that the classification accuracy of 3-D Gabor+GBDT method is higher than that of CNN algorithm,Gabor and EMAP method as texture feature,and higher than that of CNN and SVM method as classifier.Although the modeling training of 3-D Gabor+GBDT takes a long training time,the classification efficiency of this method is still high on the premise of maintaining high accuracy,which is suitable for large-scale hyperspectral image online classification scenes.