基于目标优化的高光谱图像亚像元定位

发布时间：2018-01-27 01:48

本文关键词： 二进制粒子群优化高光谱图像亚像元定位空间相关性光谱解混　出处：《中国图象图形学报》2016年06期 　论文类型：期刊论文

【摘要】：目的高光谱图像混合像元的普遍存在使得传统的分类技术难以准确确定地物空间分布,亚像元定位技术是解决该问题的有效手段。针对连通区域存在孤立点或孤立两点等特例时,通过链码长度求周长最小无法保证最优结果及优化过程计算量大的问题,提出了一种改进的高光谱图像亚像元定位方法。方法以光谱解混结合二进制粒子群优化构建算法框架,根据光谱解混结果近似估计每个像元对应的亚像元组成,通过分析连通区域存在特例时基于链码长度求周长最小无法保证结果最优的原因,提出修改孤立区域的周长并考虑连通区域个数构造代价函数,最后利用二进制粒子群优化实现亚像元定位。为了减少算法的时间复杂度,根据地物空间分布特点,采用局部分析代替全局分析,提出了新的迭代优化策略。结果相比直接基于链码长度求周长最小的优化结果,基于改进的目标函数优化后,大部分区域边界更明显,并且没有孤立1点和孤立两点的区域,识别率可以提高2%以上,Kappa系数增加0.05以上,新的优化策略可以使算法运算时间减少近一半。结论实验结果表明,本文方法能有效提高亚像元定位精度,同时降低时间复杂度。因为高光谱图像中均匀混合区域不同地物的分布空间相关性不强,因此本文方法适用于非均匀混合的高光谱图像的亚像元定位。
[Abstract]:Objective because of the prevalence of mixed pixels in hyperspectral images, it is difficult for traditional classification techniques to accurately determine the spatial distribution of ground objects. Sub-pixel localization technique is an effective method to solve this problem. When there are isolated points or isolated two points in the connected region and other special cases. Solving the minimum circumference by the length of chain code can not guarantee the optimal result and the problem that the calculation of the optimization process is heavy. An improved sub-pixel localization method for hyperspectral images is proposed. The algorithm frame is constructed by spectral demultiplexing and binary particle swarm optimization. The sub-pixel composition of each pixel is estimated approximately according to the spectral unmixing results. By analyzing the reason that the minimum circumference based on chain code length can not guarantee the optimal result when there is a special case in the connected region, a cost function is proposed to modify the circumference of the isolated region and to consider the number of connected regions to construct the cost function. Finally, binary particle swarm optimization is used to realize sub-pixel localization. In order to reduce the time complexity of the algorithm, the local analysis is used to replace the global analysis according to the spatial distribution characteristics of the objects. A new iterative optimization strategy is proposed. The results show that most of the region boundaries are more obvious than those based on the chain code length and the improved objective function. Moreover, the recognition rate can be increased by more than 2% and the Kappa coefficient can be increased by more than 2% if there is no isolated region of 1 point and isolated two points. The new optimization strategy can reduce the computational time of the algorithm by nearly half. Conclusion the experimental results show that the proposed method can effectively improve the accuracy of sub-pixel localization. Because the spatial correlation of different ground objects in homogeneous mixing region in hyperspectral images is not strong, this method is suitable for sub-pixel localization of heterogeneous hyperspectral images.
【作者单位】：杭州电子科技大学计算机应用技术研究所;浙江大学电气工程学院;
【基金】：浙江省自然科学基金项目(LZ14F030004) 国家自然科学基金项目(61571170)~~
【分类号】：TP751
【正文快照】： 0引言高光谱图像包含了从可见光到近红外几百个连续的狭窄波段光谱信息,克服了宽波段遥感探测的局限,可以对物质表面信息进行完整的光谱描述,被广泛应用于多种领域,成为对地观测重要的信息源之一。但由于成像原理与制造技术等因素的限制,高光谱图像的空间分辨率相对较低,混合

【相似文献】