多／高光谱 遥感数据的类立体纹理特征

Study on quasi-3-dimensional texture features for multi-, hyper-spectral remote sensing data analysis

摘要:

鉴于多／高光谱遥感数据同源同点多波段同时获取的特点，提出了基于灰度级差关联概率矩阵 ( Gray Level Difference Associated Possibility matrix．GLDAP)的视觉差异分析方法，以有效地利用图像底层数 据及数据之间的相关性．根据地物的波谱特性，统计两波段图像灰度协同变化的规律并记录在GLDAP矩阵 中，基于此矩阵提取了遥感数据的类立体纹理特征．将该方法与灰度共生矩阵( GLCM)纹理分析方法的遥感 地物分类性能比较，实验结果表明：基于GLDAP的纹理提取及分析表现出良好的性能，3种地物分类效果 明显优于GLCM方法，能够减少因单波段中地物可分性差而导致的误识，克服了GLCM方法对图像统计描述 的局限性，在相同时间开销下GLDAP右法较GLCM有更优的解译分析精度

Abstract:

According to the synchronous acquirement of multi-, hyper-spectral remote sensed imagery, a Gray Level Difference Associated Possibility matrix ( GLDAP) method is proposed in the paper to analyze visual differences between multi-band data. The matrix is built on two bands of image that are selected in light of land-cover spectrum characteristics. Thereafter, the co-varying statistics of gray level in each image is recor- ded and quasi-3-dimention texture features are extracted based on GLDAP. During experiments, GLDAP is employed in classifications and annotations of land cover types, compared with GLCM method. The results re- veal that the GLDAP has better performances than GLCM. Moreover, it could overcome the limitation of single band processing and understanding, on which GLCM based, and to a certain degree, decrease misrecognition rate caused by worse visual discrimination of land types at data level. The two methods have same time com- plexity; hence, GLDAP may be accepted as another choice in getting excellent precision and better perform- ance under the same time consuming