论文部分内容阅读
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。
The ubiquitous existence of hybrid hyperspectral imagery makes it difficult for the traditional classification technology to accurately determine the spatial distribution of features. The sub-pixel location technique is an effective way to solve this problem. Aiming at the problem that the optimal length of the chain length can not be guaranteed by the length of the chain code and the computational load of the optimization process is large when there are isolated points or two isolated points in the connected area, an improved sub-pixel positioning method of hyperspectral image . Methods Spectral unmixing combined with binary particle swarm optimization algorithm was used to construct the algorithm framework, and the corresponding sub-pixel composition of each pixel was estimated based on the results of spectral unmixing. By analyzing the minimum length of the chain code, The optimal reason is proposed to modify the perimeter of the isolated area and to construct the cost function considering the number of connected areas. Finally, the sub-pixel localization is achieved by binary particle swarm optimization. In order to reduce the time complexity of the algorithm, a new iterative optimization strategy is proposed based on the spatial distribution of features, using local analysis instead of global analysis. Results Compared with the optimization results based on the minimum length of chain code, the boundary of most of the regions is more obvious based on the improved objective function optimization. The recognition rate can be improved by more than 2% , The Kappa coefficient increases more than 0.05, and the new optimization strategy can reduce the algorithm operation time by nearly half. Conclusion The experimental results show that this method can effectively improve the positioning accuracy of sub-pixel and reduce the time complexity. Because the spatial distribution of different features in a homogeneous mixture region in a hyperspectral image is not strong, the proposed method is suitable for the sub-pixel localization of hyperspectral images with heterogeneous mixing.