机构地区: 中国科学院新疆生态与地理研究所
出 处: 《光谱学与光谱分析》 2011年第9期2467-2470,共4页
摘 要: 针对常规的"影像-影像"多光谱影像波段模拟模型对研究区参考影像的依赖和随机性采样带来的模型不稳定问题以及"光谱库-影像"波段模拟模型中光谱库在类别、时间和空间上的不完善导致模型缺乏适用性问题,本文提出了基于影像光谱库的波段模拟方法。首先对与待模拟影像具有相似类别组成的参考影像进行光谱聚类,然后对各个光谱类别进行采样,形成影像光谱库,接着提取光谱库中的等量类别样本训练BP神经网络(BPN)回归模型,最后利用BPN模拟目标影像的波段。实验结果表明:该方法能够更为精确的模拟出TM蓝波段,其模拟均方根误差(RMSE)较"光谱库-影像"模型提高了1.3,较"影像-影像"模型提高了0.6,与"影像-影像"和"光谱库-影像"模型相比,该方法更加稳定可靠;该方法能够成功的应用到SPOT和MSS的蓝波段的模拟中并取到较好的效果。 The authors proposed an image spectral library based band simulation method. Firstly, the authors clustered the reference image which has the same class composition with the target image by using its pixel spectrum similarity. Secondly, the authors fetched sample from the reference image base on the former cluster image, and then built the image spectral library. Thirdly, the authors fetched the same count of each type of samples to train the simulation model. Finally, the authors simulated the target band of the target image. The experiment results show that: firstly, this method can be more precise to simulate TM blue band, and increase more than 1.2 RMSE value than that of the "Spectral Library-image" model and more than 0. 6 RMSE value than that of the "image-image" model. On the other hand, our method is more stable and reliable than the "image-image" and "Spectral Library-Image" simulation model; finally, this method can be successfully applied to the blue band simulation that SPOT and MSS lacked.
领 域: [自动化与计算机技术] [自动化与计算机技术]