高光谱影像光谱-空间多特征加权概率融合分类

提出了一种基于光谱-空间多特征加权概率融合的高光谱影像分类方法。首先,利用最小噪声分离(minimum noise fraction,MNF)方法对高光谱影像进行降维和特征提取,并以得到的MNF特征影像作为光谱特征,联合灰度共生矩阵(gray level co-occurrence matrix,GLCM)提取的纹理特征、基于OFC算子建立的多尺度形态学特征以及采用连续最大角凸锥(sequential maximum angle convex cone,SMACC)提取的端元组分特征,组成3组光谱-空间特征;然后利用支持向量机(support vector machine,SVM)对每一组光谱-空间特征进行分类,得到每组特征的概率输出结果;最后,建立多特征加权概率融合模型,应用该模型将不同特征的概率输出结果进行加权融合,得到最终分类结果。为了验证该方法的有效性,利用ROSIS和 AVIRIS影像进行试验,总体分类精度分别达到97.65%和96.62%。结果表明本文的方法不但较好地克服了传统基于单一特征高光谱影像分类的局限性,而且其分类效果也优于常规矢量叠加(vector stacking,VS)和概率融合的多特征分类方法,有效地改善了高光谱影像的分类结果。

Hyperspectral Image Classification Based on the Weighted Probabilistic Fusion of Multiple Spectral-spatial Features

A hyperspectral images classification method based on the weighted probabilistic fusion of multiple spectral-spatial features was proposed in this paper.First,the minimum noise fraction (MNF)approach was employed to reduce the dimension of hyperspectral image and extract the spectral feature from the image,then combined the spectral feature with the texture feature extracted based on gray level co-occurrence matrix (GLCM),the multi-scale morphological feature extracted based on OFC operator and the end member feature extracted based on sequential maximum angle convex cone (SMACC)method to form three spectral-spatial features.Afterwards,support vector machine (SVM)classifier was used for the classification of each spectral-spatial feature separately.Finally, we established the weighted probabilistic fusion model and applied the model to fuse the SVMoutputs for the final classification result. In order to verify the proposed method,the ROSIS and AVIRIS image were used in our experiment and the overall accuracy reached 97.65% and 96.62% separately.The results indicate that the proposed method can not only overcome the limitations of traditional single-feature based hyperspectral image classification,but also be superior to conventional VS-SVM method and probabilistic fusion method.The classification accuracy of hyperspectral images was improved effectively.