空间信息自适应融合的高光谱图像分类方法

发布时间：2018-09-08 17:59

【摘要】：针对单一的滤波器提取高光谱图像空间纹理信息时不能获得完整的图像特征的不足,提出一种结合双边滤波和域转换标准卷积滤波的高光谱图像分类算法.该方法采用空间信息自适应融合的分类寻优,先对高光谱波段进行抽样分组,再用双边滤波和域转换标准卷积滤波对分组后的波段进行滤波,两种空间信息进行线性融合后交由支持向量机完成分类.实验表明,相比使用光谱信息、高光谱降维、空谱结合的支持向量机分类方法和边缘保持滤波以及递归滤波的方法,本文所提算法对高光谱图像的分类精度有较大提高,在训练样本仅为5%和3%的情况下,对印第安农林和帕维亚大学图像的总体分类精度分别达到了96.95%和97.89%,比其他算法高出213个百分点,验证了该方法在高光谱图像分类的有效性.
[Abstract]:In order to solve the problem that a single filter can not obtain complete image features when extracting spatial texture information from hyperspectral images, a hyperspectral image classification algorithm combining bilateral filtering and domain conversion standard convolution filtering is proposed. In this method, the classification and optimization of spatial information adaptive fusion are adopted. The hyperspectral bands are sampled and grouped first, and then the band after grouping is filtered by bilateral filtering and domain conversion standard convolution filtering. After linear fusion, the two spatial information are classified by support vector machine (SVM). The experimental results show that the proposed algorithm improves the classification accuracy of hyperspectral images greatly compared with the support vector machine classification method based on spectral information, hyperspectral dimensionality reduction, space-spectrum combination, edge preserving filtering and recursive filtering. When the training samples are only 5% and 3%, the overall classification accuracy of Indian agriculture, forestry and Pavia University images is 96.95% and 97.89% respectively, which is 213 percentage points higher than that of other algorithms. The validity of this method in hyperspectral image classification is verified.
【作者单位】：广东交通职业技术学院计算机工程学院;哈尔滨工程大学信息与通信工程学院;
【基金】：国家自然科学基金项目(Nos.61275010,61675051) 广东省高等职业教育品牌专业建设项目(No.2016gzpp044)资助~~
【分类号】：TP391.41

【相似文献】