联合局部二值模式的高光谱影像空-谱分类方法

Hyperspectral Imagery Spatial-Spectral Classification Combining Local Binary Patterns

为充分利用高光谱影像"图谱合一"的特性,提出了一种联合局部二值模式的高光谱影像空-谱分类方法。该方法通过局部二值模式从降维影像中提取空间纹理特征,以线性加权求和核为多核组合方式,与原始光谱特征结合构造混合核极限学习机模型,实现影像的地物分类。为了验证该方法的有效性,利用Indiana和Pavia U两组高光谱影像数据进行实验,总体分类精度分别达到99.23%和94.95%。结果表明该方法分类效果优于纯光谱分类、纯局部二值模式空间分类、GLCM空-谱分类以及3Gabor空-谱分类方法,有效地改善了高光谱影像分类结果,获得更加平滑的分类结果图。

A hyperspectral imagery spatial-spectral classification method combing local binary patterns was developed to make full use of the imagery characteristic of combination of image and spectrum. Local binary patterns approach was employed to extract spatial texture features from the dimension reduced image. Spatial and spectral kernels were integrated into a kernel framework using a linear weighted summation kernel to get a trained composite kernel extreme learning machine model acquiring the classification of image target. In order to verify the proposed method,the Indiana and Pavia U image were used in the experiment and the overall accuracies reached 99. 23%and 94.95% respectively. The results indicate that the proposed method is superior to the classification methods which use spectral features,spatial features of local binary patterns or spatial-spectral features based on GLCM and3 Gabor. The classification accuracy of hyperspectral imagery is improved effectively and the smoother classification map can be obtained.