基于脉冲耦合神经网络的高光谱遥感图像融合方法研究

The space-based hyperspectral remote sensing images have rich spectral details, whereas they inevitably have the shortcomings of low resolution. Therefore, one possible solution comes from taking advantages of panchromatic or SAR images fusion, which will result in the improvement of remote sensing information analysis. However, there are some problems need to be solved in order to achieve this goal. Spectral region of panchromatic images is narrow, which leads to the false spatial details of fusion result. In addition, The image registration is difficult to implement due to the large difference among the multi-source remote sensing images. At the same time, spectral fidelity is rarely considered by the classical methods in the SAR image fusion. Lastly, the existing fusion algorithm mostly focuses on the geometric distribution of pixel values, which ignores application of the nonlinear characteristics of human eyes for the hyperspectral fusion. Hence, the project intends to improve the pulse coupled neural network (PCNN) model which conforms to the human visual characteristics, and propose multi-source remote sensing image fusion algorithms through studying the multi-source remote sensing imaging mechanism. The main contents are as follows. (1) Hyperspectral pansharpening approach using PCNN. (2) Fusion approach of high resolution SAR and hyperspectral images. (3) Validation and analysis. Through the implementation of this project, it is expected to improve the spectral resolution and space resolution of fusion images, and obtain the hyperspectral fusion results with SAR characteristics. The proposed methods can also provide key algorithms and theoretical support from a new perspective for the hyperspectral remote sensing fusion.

高光谱星载遥感图像光谱分辨率高，但其空间分辨率低。若采用全色和高分SAR图像与其进行融合，可大大提高后续遥感信息对地物细节的解译能力。对其融合尚需进一步解决全色图像光谱覆盖窄引起的空间细节误匹配、多源大尺度遥感图像配准困难、与SAR融合的光谱保真等问题。此外，现有融合算法多受限图像本身几何特性理解，鲜有将人眼视觉非线性特性真正应用到高光谱图像融合。为此，本项目拟结合多源遥感成像机理，改进符合人眼视觉特性的脉冲耦合神经网络模型，开展高光谱图像与全色图像及高分SAR图像的融合方法研究，主要内容包括：（1）基于改进脉冲耦合神经网络模型的全色图像与高光谱图像融合算法设计；（2）高分SAR与高分辨高光谱图像的融合算法设计；（3）验证与分析。通过本项目实施，可望在提高融合图像的光谱分辨率和空间分辨率同时，获得具有SAR散射特征的高光谱融合结果，可从新的角度为高光谱融合研究提供关键算法和理论支持。

高光谱遥感图像的优点是光谱波段较多，但是由于光学遥感传感器信噪比的约束，使得高光谱图像受低空间分辨率影响，无法满足其精细化的解译应用需求。脉冲耦合神经网络（PCNN）是一种非训练生物型反馈神经网络模型，其神经元间同步脉冲发放特性可以非线性分割图像、提取图像的不变特征，有益于图像融合。因此，本项目基于PCNN模型，开展多源遥感图像融合方法研究，主要成果包括：（1）提出了基于PCNN模型点火集群约束的遥感图像配准算法，降低了融合前特征点误配现象，提高了配准精度；（2）实现了基于PCNN模型非线性分割、非规则细节注入的多、高光谱全色锐化算法，提出了基于交叉皮质神经网络模型的SAR图像融合算法，增加了融合图像的散射信息；（3）实现了自适应参数优化的PCNN多源遥感图像融合算法，解决了PCNN模型用于图像融合过程中需要人工设置参数问题；（4）建立了专门用于全色图像锐化融合的PCNN改进模型和改进的交叉皮质神经网络模型，降低了融合图像光谱畸变，提升了融合图像空间细节；（5）提出了结合PCNN模型的无人机航片与高分辨率卫星图像融合算法，获得了亚米级分辨率的融合图像。本项目的研究成果减少了遥感图像融合中的光谱失真，增加了融合结果的空间细节，为PCNN等视觉皮层神经元模型在多源遥感图像融合的研究提供了理论依据。