高光谱图像去噪和目标检测一体化处理方法研究

Target detection is one of important problems on hyperspectral image processing. When detecting target from noisy hyperspectral image, the conventional process is detecting target a after denoising. Because denoising processing often results in destruction of hyperspectral image information which will reduce the target detection performance. For this, the integration method for hyperspectral image denoising and target detection is proposed, in which denoising and target detection are processed side by side for denoising and detecting target simultaneously. Firstly, a dimensionality reduction algorithm is proposed for reducing the redundancy of hyperspectral image and the computation cost, and for providing effective data for the integrated model as well. Secondly, the integrated model is studied in which some models are included: (1) A model for linking the processes of denoising and target detection is studied based on Low-rank representation of noise-free background image. (2) A constraint model for joint processing of random noise and sparse noise is studied. (3) A constraint model for target detection is studied. (4) By using multitask learning theory, the integration objective function is constructed based on the linking model, the constraint model of denoising and the constraint model of target detection. The research of our program can provide a theoretical method and a key technical support for application and development of hyperspectral image.

对存在噪声的高光谱图像进行目标检测时，常规的处理方式是“先去噪后目标检测”，然而去噪处理往往会造成图像信息的破坏，从而影响目标的正确检测。鉴于此，本项目研究一种高光谱图像去噪和目标检测一体化处理方法，旨在通过“边去噪，边目标检测”的处理方式，达到在去噪的同时正确检测目标的目的。首先，开展高光谱图像数据降维算法的研究，以通过数据降维，减少信息的冗余和数据处理计算量，同时为一体化模型的建立提供更有效的数据。其次，重点研究一体化处理模型，具体包括：（1）从对无噪声背景图像进行低秩表示的角度研究去噪和目标检测处理的衔接模型；（2）研究随机噪声和稀疏噪声联合处理的约束模型；（3）研究目标检测约束模型；（4）基于衔接模型、去噪处理约束模型和目标检测约束模型，结合多任务学习理论，构建去噪和目标检测一体化处理目标函数。本项目的研究可为高光谱图像的广泛应用提供理论方法和技术支持。

目标检测是高光谱图像处理中的一个重要问题。针对存在噪声情况下的高光谱图像目标检测问题，本项目研究了一种高光谱图像去噪和目标检测一体化处理方法，旨在通过“边去噪，边目标检测”的处理方式，达到在去噪的同时正确检测目标的目的。首先，研究了高光谱图像数据降维算法，以通过数据降维，减少信息的冗余和数据处理计算量，同时为一体化模型的建立提供更有效的数据。其次，重点研究了一体化处理模型，具体包括：研究了高光谱图像去噪处理模型、光谱增强模型、目标检测模型、去噪和目标检测一体化处理模型。最后，利用收集的高光谱图像分别进行各模型的有效性验证。本项目的研究可为高光谱图像的广泛应用提供理论方法和技术支持。