顾及邻域像元分类决策的遥感影像分类蚁群算法

发布时间：2018-05-16 16:44

本文选题：遥感影像分类 + 蚁群规则挖掘　；参考：《测绘地理信息》2016年06期

【摘要】：为了进一步发掘蚁群算法的应用潜力,提高分类精度,将相关性引入分类规则发掘过程,试图在蚁群算法挖掘规则时既考虑像元的光谱信息,又兼顾邻近像元灰度的空间相关性,提出了一种优化的蚁群算法。算法包括对单个像元的分类规则挖掘和顾及邻域像元相关性的分类规则挖掘,单个像元的分类规则挖掘中,为使信息素缓和增加,避免陷入局部最优解,同时保证算法具有适当的收敛速度,采用自适应方案调整参数。顾及邻域像元相关性的分类规则挖掘中选用了优势类、优势度、类熵和邻域类相关性等4个指标,以反映邻域相关性对分类结果的影响。实验研究发现,顾及邻域蚁群算法的分类结果精度有了较为明显的提高,总体精度提高了3.00%,其优势主要体现在对建设用地、裸地等复杂地物的识别。研究结果表明,顾及邻域蚁群算法能够更准确地提取光谱信息复杂的地物,有效地减弱同物异谱和异物同谱现象的干扰。
[Abstract]:In order to further explore the application potential of ant colony algorithm and improve classification accuracy, correlation is introduced into the process of classification rules mining, and the spectral information of pixel is taken into account when mining rules in ant colony algorithm. Considering the spatial correlation of adjacent pixels, an optimized ant colony algorithm is proposed. The algorithm includes classification rules mining for single pixel and classification rule mining considering neighborhood pixel correlation. In single pixel classification rule mining, pheromones can be added to avoid falling into local optimal solution. At the same time, the adaptive scheme is used to adjust the parameters to ensure that the algorithm has a proper convergence rate. In order to reflect the influence of neighborhood correlation on classification results, four indexes are selected in the mining of classification rules that take into account the correlation of neighborhood pixel, such as dominant class, dominance degree, class entropy and neighborhood class correlation. The experimental results show that the classification accuracy of considering the neighborhood ant colony algorithm has been improved obviously and the overall accuracy has been improved by 3.00. The advantages of the algorithm are mainly reflected in the identification of complex ground objects such as construction land bare land and so on. The results show that considering the neighborhood ant colony algorithm, the complex objects with complex spectral information can be extracted more accurately, and the interference of isospectral and foreign body isospectral phenomena can be effectively reduced.
【作者单位】：湖南省第一测绘院;衡阳师范学院城市与旅游学院;湖南省古村古镇文化遗产数字化传承协同创新中心;
【基金】：国家自然科学基金资助项目(41471118) 教育部人文社科资助项目(11YJZH255)~~
【分类号】：P237

【相似文献】