【作者】 吴丹；

【导师】 冯收；

【作者基本信息】 哈尔滨工程大学 ， 信息与通信工程， 2023， 硕士

【摘要】 高光谱图像目标检测在军事和民用应用中都有重要的作用。根据是否可以获知目标的光谱先验信息,目标检测技术可以被分为异常目标检测技术和特定目标检测技术。由于异常目标检测任务缺乏目标的先验光谱信息,在对背景进行建模和估计时很难完全排除异常目标及噪声的干扰,从而削弱模型对背景的表达能力。在高光谱图像特定目标检测中,虽然可以获知目标的光谱信息,但是由于高光谱图像波段间信息冗余度高,加之噪声等因素的影响,使得目标和背景在光谱信息上的区分度变得不明显,从而导致检测方法的结果精度不理想。针对上述问题,本文分别从异常目标检测和特定目标检测两个方面展开研究。1.在未知目标先验信息情况下,针对异常目标及噪声的干扰而降低背景信息获取和建模准确性的问题,本文提出了一种基于全变差正则化低秩张量分解和协同表示的异常检测方法。该方法利用全变差正则化低秩张量分解模型将高光谱图像分解成混合信息和低秩信息两部分,然后结合两个信息部分的特点分别设计异常检测器来提取异常信息。最后,通过对两部分的异常检测结果加权融合,得到最终的异常目标检测结果。本文通过在三个真实的高光谱数据集上进行检测,并与其他8种异常目标检测方法进行比较,验证了该方法的有效性,该方法在三个数据集上的异常目标检测精度分别为0.9954、0.9804和0.9869。2.在已知目标先验信息情况下,针对因高光谱图像高冗余性及噪声等因素而降低目标和背景区分度的问题,本文提出了一种基于双因子正则化低秩张量分解的双阶段特定目标检测方法。该方法采用双因子正则化低秩张量分解模型来提取背景张量,降低噪声因素的影响。然后,采用由粗到精的策略来逐步增加目标和背景的区分度,得到最终的特定目标检测结果。本文通过在三个真实的高光谱数据上进行检测,并与其他8种特定目标检测方法进行比较,验证了该方法的有效性,该方法在三个数据集上的特定目标检测精度分别为0.9976、0.9819和0.9987。更多还原

【Abstract】 Hyperspectral image target detection plays an important role in military and civilian applications.According to whether the spectral prior information of the target can be obtained,the target detection technology can be divided into anomalous target detection and specific target detection.Because the anomalous target detection task lacks the prior spectral information of the target,it is difficult to completely eliminate the interference of anomalous target and noise when modeling and estimating the background,thus weakening the model ’s ability to express the background.In the specific target detection of hyperspectral image,although the spectral information of the target can be obtained,due to the high information redundancy between the bands of hyperspectral image and the influence of noise and other factors,the discrimination between the target and the background in the spectral information becomes not obvious,which leads to the unsatisfactory accuracy of the detection method.In view of the above problems,this paper studies from two aspects: anomalous target detection and specific target detection.1.In the case of unknown target prior information,an anomalous target detection method based on total variation regularized low-rank tensor decomposition and collaborative representation is proposed to solve the problem of reducing the accuracy of background information acquisition and modeling due to the interference of anomalous target and noise.The method uses the total variation regularized low-rank tensor decomposition model to decompose the hyperspectral image into two parts: mixed information and low-rank information,and then combines the characteristics of the two information parts to design anomalous detector to extract anomalous information.Finally,the final anomalous target detection result is obtained by weighted fusion of the two parts of the anomalous detection results.In this paper,the effectiveness of the method is verified by testing on three real hyperspectral datasets and comparing with eight other anomalous target detection methods.The anomalous detection accuracy of this method on the three datasets is 0.9954,0.9804 and0.9869,respectively.2.In the case of known target prior information,a two-stage specific target detection method based on double-factor regularized low-rank tensor decomposition is proposed to solve the problem of reducing target and background discrimination due to high redundancy and noise of hyperspectral images.The method uses the double-factor regularized low-rank tensor decomposition model to extract the background tensor and reduce the influence of noise factors.Then,a coarse-to-fine strategy is used to gradually increase the discrimination between the target and the background to obtain the final specific target detection result.In this paper,the effectiveness of the method is verified by testing on three real hyperspectral datasets and comparing with eight other specific target detection methods.The specific target detection accuracy of this method on the three datasets is 0.9976 、 0.9819 and 0.9987,respectively.更多还原