一种新的用于高光谱图像小目标探测的目标光谱学习算法

发布时间：2018-03-25 03:14

本文选题：高光谱图像　切入点：目标探测　出处：《红外与毫米波学报》2017年04期

【摘要】：提出一种用于高光谱图像小目标探测的目标光谱学习算法,目的是从图像中学习得到一条更为准确的目标光谱,从而提高有监督目标探测的效果.该算法由基于自适应权重的目标光谱学习算法和自完备字典两部分组成.前一部分内容是在已有完备的背景字典的情况下,通过稀疏编码和梯度下降算法来优化学习目标光谱;后一部分内容通过背景字典的不断扩充来确保该字典的完备性,从而保证了学习算法的准确性.仿真和实际高光谱数据的实验结果表明,所提出的方法能有效地提取出准确的目标光谱,从而显著提高目标探测算法的精度.
[Abstract]:Put forward a target for the spectral learning algorithm for small target detection of hyperspectral image from the image, to learn a more accurate target spectrum, thereby improving the supervised target detection results. The algorithm consists of target spectral learning algorithm based on adaptive weight and complete dictionary is composed of two parts. The first part in the background there is complete dictionary of the case, to optimize the learning target spectrum by the sparse encoding and gradient descent algorithm; part after by expanding background to ensure the completeness of the dictionary from the dictionary, and ensure the accuracy of the learning algorithm. The simulation and experimental results on real hyperspectral data show that the proposed method can effectively extract the target spectrum accurately, so as to improve the accuracy of target detection algorithm.

【作者单位】：复旦大学电磁波信息科学教育部重点实验室;复旦大学信息学院智慧网络与系统研究中心;
【基金】：国家自然科学基金(61572133) 北京师范大学地表过程与资源生态国家重点实验室开放基金(2015-KF-01)~~
【分类号】：TP751

【相似文献】