基于烟花算法降维的高光谱图像分类

发布时间：2018-05-30 22:44

本文选题：图像分类 + 高光谱图像　；参考：《华南理工大学学报(自然科学版)》2017年03期

【摘要】：为降低高光谱数据量及计算复杂度,避免后续分类中的Hughes现象,提出一种基于烟花算法降维的高光谱图像分类方法.烟花算法采用类内紧密性系数与类间分离性系数的加权和作为波段选择的度量准则,通过在高光谱数据空间内进行搜索,不断更新直至收敛,从而获得最优波段组合.基于印第安纳数据集(AVIRIS)和帕维亚大学数据集(ROSIS)数据对烟花算法、遗传算法和禁忌搜索算法进行波段选择的仿真实验,结果表明:烟花算法选择出的波段组合数目相对较少,具有较低的算法复杂度,减少了耗时;利用获得的波段组合进行高光谱图像分类时,与遗传算法、禁忌搜索算法的波段选择方法相比,文中所提方法在总体分类精度和Kappa系数上分别提高0.06%~4.72%和0.00~0.09,可以得到令人满意的分类结果.
[Abstract]:In order to reduce the amount of hyperspectral data and the computational complexity and avoid the Hughes phenomenon in the subsequent classification, a hyperspectral image classification method based on reducing the dimension of the fireworks algorithm is proposed. In the fireworks algorithm, the weighted sum of intra-class compactness coefficient and inter-class separation coefficient is used as the criterion of band selection. By searching in hyperspectral data space, the algorithm is updated and converged to obtain the optimal band combination. Based on the data of Indiana data set AVIRISand Pavia University data set Rosis, the simulation experiments of band selection of fireworks algorithm, genetic algorithm and Tabu search algorithm are carried out. The results show that the number of band combinations selected by the fireworks algorithm is relatively small. Compared with genetic algorithm and Tabu search algorithm, the band selection method for hyperspectral image classification based on the obtained band combination is less time consuming and lower complexity than that of genetic algorithm and Tabu search algorithm. The method proposed in this paper can improve the classification accuracy and Kappa coefficient by 0.06% and 0.000. 09%, respectively. Satisfactory classification results can be obtained.
【作者单位】：哈尔滨工程大学信息与通信工程学院;黑龙江省农业科学院遥感技术中心;
【基金】：国家自然科学基金资助项目(61675051) 教育部博士点基金资助项目(20132304110007) 黑龙江省博士后特别资助项目(LBH-TZ0420)~~
【分类号】：TP18;TP391.41

【相似文献】