|
发布时间: 2018-11-16 |
NCIG 2018会议专栏 |
|
|
收稿日期: 2018-04-17; 修回日期: 2018-06-05
基金项目: 国家自然科学基金项目(U1736217)
第一作者简介:
荣楚君, 1994年生, 男, 硕士研究生, 主要研究方向为红外小目标检测与跟踪。E-mail:459187614@qq.com;
曹晓光, 男, 副教授, 主要研究方向为图像模式识别应用。E-mail:xgcao@buaa.edu.cn.
中图法分类号: TP391
文献标识码: A
文章编号: 1006-8961(2018)11-1768-09
|
摘要
目的 红外弱小目标检测是红外图像处理领域中难度大且实际意义相当重要的一项研究热点问题,其在侦察预警系统、飞行器跟踪系统与导弹制导系统中都扮演了十分重要的角色。自然背景下的红外图像一般具有较低信噪比,其中背景占据着绝大部分面积,而目标尺寸很小且不具有明显形状和纹理信息,这为红外图像中弱小目标的检测增加了难度。本文提出一种将Facet方向导数特征与稀疏表示相结合的红外弱小目标检测算法。方法 首先利用Facet模型提取原红外图像在0°、90°、45°和-45° 4个方向上的一阶导数特征,然后通过稀疏表示方法,在方向导数信息基础上对图像进行分块逐一处理,利用求解出的稀疏系数和导数图像块的重建残差构建检测数值图,最后分割出小目标所在具体位置。结果 通过对4组不同红外图像序列进行实验验证,绘制了检测率与虚警率ROC曲线图。从结果可以看出,本文算法相较于对比算法在小目标检测中具有较高检测率。结论 本文算法将Facet方向导数特征与稀疏表示相结合,在红外弱小目标检测上具有较高检测精度和较强抗噪声干扰能力,相比于传统检测算法具有一定优势,同时可根据不同检测背景训练出相应背景字典,从而得到较好检测效果,在实际工程应用中具有良好针对性。
关键词
红外图像; 目标检测; 小目标; 方向导数; 稀疏表示
Abstract
Objective Infrared dim and small target detection is a research interest in the field of infrared image processing, which is difficult but practical. It plays a crucial role in reconnaissance and warning, aircraft tracking, and missile guidance systems. The process of detecting infrared small targets in natural scenes is characterized by the fact that the target area can frequently be expressed as a small, uniform, compact area with a significant discontinuity or contrast compared with the surrounding background. The detection of a small target in an infrared image is affected by many factors, such as the small number of target pixels, low contrast between a target and a background, dim edges of the targets, complex image background, and lack of texture information of the small targets, thereby resulting in the difficulty of infrared small target detections. The existing methods have achieved effective results in detecting small targets in infrared images; however, drawbacks, such as low adaptability to complex background, low detection rate, and high false alarm rate, still remain. In addition, methods related to sparse representation have the following shortcomings:the construction of a dictionary directly from the original images ignores the feature extraction of the target, or does not establish the target and the background dictionaries simultaneously, thus resulting in a weak representation capability of the entire dictionary. Thus, an infrared small target detection algorithm that combines facet directional derivative features with sparse representation is proposed. Method A dictionary must initially be constructed. A background dictionary is constructed by intercepting 1 000 small blocks and then obtaining their derivatives in a certain direction. K-SVD algorithm is used to train the blocks after merging them into column vectors. A background dictionary with 500 atoms is achieved. The construction method of the target dictionary is as follows:325 small blocks containing small targets are generated in accordance with the characteristics of the small target. The first-order derivative in one direction is calculated for these small blocks containing small targets, and then the columns are converted into column vectors. The target dictionary containing 325 atoms is obtained in that direction after normalizing. We combine the target and the background dictionaries into one large dictionary with 825 atoms, which will be used in the subsequent sparse solution section. The facet model is utilized to extract the first-order derivative features of the original infrared image in four directions, that is 0°, 90°, 45°, and -45°. Then, the blocks separated from the image are processed from top to bottom and left to right on the basis of the directional derivative information through the sparse representation method. The detection result map is constructed using the sparse coefficients and reconstruction residuals of the derivative image blocks. Finally, a threshold is calculated from the detection result map to separate the target from the background. Result The classical max-mean and max-median algorithms are selected as the algorithms for comparison. Comparative results show that the max-mean and max-median algorithms are sensitive to the edges in the infrared image. The traditional algorithms perform ineffectively in removing these clusters when the infrared image has clusters due to distance, atmospheric refraction, lens aberration, and optical defocus. A 3D image of the detection result shows that our method has better performance, is insensitive to noise, and can achieve an excellent target detection effect. Therefore, our algorithm has certain advantages over the traditional algorithms. Receiver operating characteristic (ROC) curves of detection and false alarm rates are plotted through experimental verification of four infrared image sequences. The results evidently show that the proposed algorithm has a higher detection rate and lower false alarm rate in a small target detection than other algorithms. Conclusion Our algorithm extracts image directional derivative information through the facet model, combines the directional derivative features of infrared imagery with sparse representation theory, analyzes the characteristics of the small target in a single direction in detail, and extends it to feature information presented in multiple directions. The difference between the target and the background is discussed. The final test results of the small target are obtained using sparse representation theory as a medium. Experiments show that the proposed algorithm has a high detection accuracy and strong anti-noise capability. The proposed algorithm has certain advantages, improves detection rate, and reduces false alarm rate over the traditional detection algorithms. Another important advantage of our algorithm is that it can generate different background dictionaries in accordance with a certain background under different conditions to obtain improved detection results and perform an effective pertinence in practical applications.
Key words
infrared image; object detection; small target; directional derivative; sparse representation
0 引言
红外弱小目标识别技术在军事和民用领域中都扮演着十分重要的角色。在自然场景中识别红外小目标这一过程的特点是,目标区域通常可以表述为一个尺寸小、均匀、紧凑的区域,与周围背景相比,具有显著不连续性或者强烈对比性。在一幅红外图像中检测小目标,会受到目标像素数量少、目标和背景之间对比度较低、目标边缘暗淡、图像背景复杂以及小目标纹理信息缺失等因素的影响。
红外弱小目标检测近年来被学者们广泛研究。在早期方法中,Deshpande等人[1]提出了最大均值滤波和最大中值滤波的方法用来检测红外图像中的弱小目标,Zeng等人[2]提出了顶帽形态学滤波器以及在红外目标检测中的应用,这些方法都着力于设计一种滤波器来抑制背景,从而提取小目标。Gu等人[3]提出了一种基于内核的非参数回归方法,用于红外图像中背景预测和小目标检测,Dong等人[4]提出了一种用于红外点目标检测的均匀背景预测模型,这些算法同样致力于背景抑制工作,但是都对复杂背景噪声颇为敏感。之后,Qi等人[5]设计了一种基于二阶方向导数和布尔图视觉理论的小目标检测算法,这种方法在检测率上有了很大提升,但是在应对复杂背景中小目标检测时会有较高虚警率。近几年,稀疏表示理论在信号处理与分析的领域中有着出色的表现,目标样本与背景样本的有效利用在目标检测上呈现出优良效果[6],Wang等人[7]根据稀疏表示理论提出了一种利用背景字典分块表示原红外图像方法来检测小目标位置的算法,Wan等人[8]根据稀疏表示理论提出了一种利用目标字典分块表示原红外图像从而检测出小目标位置的算法。
现有方法在红外图像中的小目标检测上已经有了较好结果,但是也存在着对复杂背景适应性低、检测率较低、虚警率较高等问题,而现有稀疏相关方法存在的问题在于:直接从原图上入手构建字典,没有考虑目标的特征提取,或者没有同时构建目标和背景字典,字典的表示能力较弱。针对这些缺陷,本文提出了一种利用红外图像的Facet方向导数特征与稀疏表示理论相结合来检测弱小目标的方法。
1 Facet模型
Haralick提出[9]:一个图像中任意一个像素点某一个邻域所有像素点灰度值所形成的灰度强度表面,都可以被一个空间的二元三次多项式拟合,现在假定这个映射函数为
$ \mathit{\boldsymbol{R}} \times \mathit{\boldsymbol{C}} = \left\{ \begin{array}{l} 1,r,c,{r^2} - 2,rc,{c^2} - 2,\\ {r^3} - 17/5r,\left( {{r^2} - 2} \right)c,\\ r\left( {{c^2} - 2} \right),{c^3} - 17/5c \end{array} \right\} $ | (1) |
令
$ \begin{array}{*{20}{c}} {f\left( {r,c} \right) = {K_1} + {K_2}r + {K_3}c + {K_4}\left( {{r^2} - 2} \right) + }\\ {{K_5}rc + {K_6}\left( {{c^2} - 2} \right) + {K_7}\left( {{r^3} - 17/5r} \right) + }\\ {{K_8}\left( {{r^2} - 2} \right)c + {K_9}r\left( {{c^2} - 2} \right) + }\\ {{K_{10}}\left( {{c^3} - 17/5c} \right)} \end{array} $ | (2) |
式中,
在本文方法中需要求解图像中任意一个像素点0°、90°、45°和-45°方向的一阶导数,如图 1所示。
$ {{f'}_0}\left| {_{\left( {0,0} \right)}} \right. = {K_3} - 2{K_8} - 3.4{K_{10}} $ | (3) |
$ {{f'}_{90}}\left| {_{\left( {0,0} \right)}} \right. = {K_2} - 2{K_9} - 3.4{K_7} $ | (4) |
$ \begin{array}{*{20}{c}} {{{f'}_{45}}\left| {_{\left( {0,0} \right)}} \right. = {{f'}_0}\left| {_{\left( {0,0} \right)}} \right. \times \cos \left( {{{45}^ \circ }} \right) + }\\ {{{f'}_{90}}\left| {_{\left( {0,0} \right)}} \right. \times \sin \left( {{{45}^ \circ }} \right)} \end{array} $ | (5) |
$ \begin{array}{*{20}{c}} {{{f'}_{ - 45}}\left| {_{\left( {0,0} \right)}} \right. = {{f'}_0}\left| {_{\left( {0,0} \right)}} \right. \times \cos \left( { - {{45}^ \circ }} \right) + }\\ {{{f'}_{90}}\left| {_{\left( {0,0} \right)}} \right. \times \sin \left( { - {{45}^ \circ }} \right)} \end{array} $ | (6) |
利用式(3)—(6)求出原图像的4个方向上的方向导数特征。
2 目标检测算法
2.1 单一方向目标检测
在自然场景中,目标区域通常可以表述为一个小尺寸、均匀、紧凑的区域,没有明显形状尺寸和纹理信息,与周围背景相比,具有显著不连续性或者强烈对比性。在通常情况下,一幅灰度图中小目标可以看做一个2维高斯函数模型[10]
$ I\left( {x,y} \right) = A\exp \left\{ { - \frac{1}{2}\left[ {{{\left( {\frac{x}{{{\sigma _1}}}} \right)}^2} + {{\left( {\frac{y}{{{\sigma _2}}}} \right)}^2}} \right]} \right\} $ | (7) |
式中,
假如现有一个从原图像中截取的尺寸为
$ \begin{array}{*{20}{c}} {\min {{\left\| {{\mathit{\boldsymbol{X}}_t}} \right\|}_0} + {{\left\| {{\mathit{\boldsymbol{X}}_b}} \right\|}_0}}\\ {{\rm{s}}.\;{\rm{t}}.\;\;{\mathit{\boldsymbol{D}}_{\rm{t}}}{\mathit{\boldsymbol{X}}_{\rm{t}}} + {\mathit{\boldsymbol{D}}_{\rm{b}}}{\mathit{\boldsymbol{X}}_{\rm{b}}} = \mathit{\boldsymbol{Y}}} \end{array} $ | (8) |
式中,
然而,在实际红外图像中,背景是非常复杂的,实际红外图像中背景部分可能包含:“斜坡”、“角点”、“悬崖”、“山谷”等,并不会形如理想情况下在某一个方向上一阶导数近似于一个平面,这些复杂背景成分会对目标检测造成不同程度的干扰[11]。尽管如此,由于目标可以被看做一个2维高斯函数模型,它在各个不同方向上都能近似为图 2(b)中所示曲线分布,这是目标在导数域上的独有特性[12],如图 3所示,而上述背景成分均不可能在各个不同方向上都呈现出近似于形如目标曲线的分布,利用这一特性,可以用Facet模型求解原图像各个方向上的一阶导数,分别求解稀疏系数,再进行多方向上融合处理,即可将目标和背景彻底区分开。
2.2 多方向目标检测
基于上述分析,本文提取多方向导数特征,通过多方向共同求解,应对实际红外图像中背景复杂的问题。本节首先介绍字典的获取与构建方式,然后详细阐述多方向求解的方法,最后说明图像重建和目标检测的方法。
2.2.1 字典获取
字典获取分为背景字典获取和目标字典获取两个部分。
对于背景字典,在图像库中不包含目标的背景区域里,截取1 000个尺寸为16×16像素的小块[13],对这1 000个小块分别利用Facet模型求4个方向上的一阶导数,然后分别按列合并得到4个256维列向量,再对它们进行归一化,由此在每一个方向上都可以得到1 000个背景样本。
为了求解出最稀疏的稀疏系数,缩短求解时间,同时还要为输入信号的稀疏表示搭建出一个很好的过完备字典,本文选择了平稳有效的K-SVD算法[14]对背景样本进行训练。K-SVD算法效率高,而且针对图像复原和图像压缩具有较好应用效果,K-SVD算法是一种通过迭代来实现字典训练的方法,在一个总循环中,主要包含稀疏编码和字典更新两个步骤。
利用K-SVD算法训练某一方向上1 000个背景样本,得到该方向上背景字典
对于目标字典的搭建,首先考虑目标小块生成过程,红外图像中小目标的特点主要包含以下3个方面:1)小目标是呈高斯分布的圆形或者椭圆形的连续、均匀、紧凑的亮区域;2)针对椭圆形小目标,有时还具有一定旋转角度;3)目标所占总像素数一般是2×2到8×8[15]。根据小目标所具有的特点一共生成325个含有小目标的小块,在这些目标块中,圆形小目标所具有的直径包括4~16的所有整数,椭圆形小目标的长轴包括5~16的所有整数,短轴包括从4到对应长轴长度减1的所有整数。此外,椭圆形小目标在0°基础上还分别旋转了45°、90°和135°。对这些含有小目标的小块求某一方向一阶导数,再按列并成256维列向量,然后对它们进行归一化,得到该方向上含有325个原子的目标字典
2.2.2 多方向求解
引入多方向对小目标检测具有很好的效果[16],本文取0°、90°、45°和-45°方向这4个方向进行求解。现对式(8)进行改进,令大字典
$ \begin{array}{*{20}{c}} {\min {{\left\| \mathit{\boldsymbol{X}} \right\|}_0}}\\ {{\rm{s}}.\;{\rm{t}}.\;\;\mathit{\boldsymbol{DX}} = \mathit{\boldsymbol{Y}}} \end{array} $ | (9) |
在
$ \begin{array}{*{20}{c}} {\min \sum\limits_{i = 1}^n {{{\left\| {{\mathit{\boldsymbol{X}}_i}} \right\|}_0}} }\\ {{\rm{s}}.\;{\rm{t}}.}\\ {\left[ {{\mathit{\boldsymbol{Y}}_1}{\mathit{\boldsymbol{Y}}_2} \cdots {\mathit{\boldsymbol{Y}}_n}} \right] = \left[ {{\mathit{\boldsymbol{D}}_1}{\mathit{\boldsymbol{D}}_2} \cdots {\mathit{\boldsymbol{D}}_n}} \right] \cdot }\\ {\left[ {\begin{array}{*{20}{c}} {{\mathit{\boldsymbol{X}}_1}}&0& \cdots &0\\ 0&{{\mathit{\boldsymbol{X}}_2}}& \cdots &0\\ \vdots&\vdots &{}& \vdots \\ 0&0& \cdots &{{\mathit{\boldsymbol{X}}_n}} \end{array}} \right]} \end{array} $ | (10) |
令
$ \underline {\mathit{\boldsymbol{\bar X}}} = \left[ {\begin{array}{*{20}{c}} {{\mathit{\boldsymbol{X}}_1}}&0& \cdots &0\\ 0&{{\mathit{\boldsymbol{X}}_2}}& \cdots &0\\ \vdots&\vdots &{}& \vdots \\ 0&0& \cdots &{{\mathit{\boldsymbol{X}}_n}} \end{array}} \right] $ | (11) |
则式(10)可简化为
$ \min \sum\limits_{j = 1}^n {{{\left\| {\underline {\mathit{\boldsymbol{\bar X}}} {._j}} \right\|}_0}} \;\;\;\;{\rm{s}}.\;{\rm{t}}.\;\;\underline {\mathit{\boldsymbol{\bar Y}}} = \underline {\mathit{\boldsymbol{\bar D}}} \cdot \underline {\mathit{\boldsymbol{\bar X}}} $ | (12) |
式中,
2.2.3 图像重建
将原红外图像从上至下,从左至右,以步长为4截取16×16像素的图像小块,对它们求0°、90°、45°和-45° 4个方向上一阶导数,利用式(12)求解稀疏系数矩阵
$ {\gamma _i} = \left\| {{\mathit{\boldsymbol{Y}}_i} - {\mathit{\boldsymbol{D}}_{{\rm{b}}i}}{\mathit{\boldsymbol{X}}_{{\rm{b}}i}}} \right\|_2^2 $ | (13) |
利用目标系数
$ {\mathit{\boldsymbol{G}}_i} = \sum\limits_{k = 1}^K {{\mathit{\boldsymbol{X}}_{{\rm{t}}i}}\left( k \right) \cdot \mathit{\boldsymbol{B}}\left( k \right)} $ | (14) |
式中,
$ \mathit{\boldsymbol{P}} = \frac{1}{n}\sum\limits_{i = 1}^n {{\gamma _i} \cdot {\mathit{\boldsymbol{G}}_i}} $ | (15) |
最后利用所有重建小块重建原图像,小块重叠部分采用均值滤波方式取值,由此得到最终检测结果图。
2.2.4 目标检测
针对最终得到的重建图像,即检测结果图,取一个阈值
$ T = \bar I + q \cdot {\sigma _r} $ | (16) |
式中,
3 实验结果与分析
本文算法的流程图如图 4所示。
3.1 ROC曲线
ROC(receiver operating characteristic)曲线是衡量目标检测结果的一个常用指标,能够反映检测率
$ DR = \frac{{{T_{{\rm{NC}}}}}}{{{T_{{\rm{NT}}}}}};FA = \frac{{{P_{{\rm{NF}}}}}}{{{P_{{\rm{NA}}}}}} $ | (17) |
以虚警率
3.2 定量实验
为了验证算法的有效性,用4组红外图像序列,第1组图像分辨率为768×576像素,一共60帧,60个目标,第2组图像分辨率为768×576像素,一共60帧,60个目标,第3组图像分辨率为768×576像素,一共60帧,60个目标,第4组图像分辨率为768×576像素,一共60帧,60个目标。4组图像序列一共240张图。
本文选择经典的Max-mean和Max-median算法作[1]为对比算法[18],每种算法的检测结果如图 5所示。其中,第1行是分别从4组图像序列中随机抽取的一张红外图像原图,第2行是Max-mean算法对这4张红外图像滤波后的结果图,第3行是Max-median算法对这4张红外图像滤波后的结果图,第4行是本文算法对这4张红外图像处理后的结果图。从图 6中可以更加清晰地看到每种算法的检测结果。其中,第1行是这4张红外图像的3维图,第2行是Max-mean算法对这4张红外图像滤波后的3维图,第3行是Max-median算法对这4张红外图像滤波后的3维图,第4行是本文算法对这4张红外图像处理后的3维图。可以看出,Max-mean和Max-median算法对红外图像中的边缘非常敏感,第3张和第4张红外图像在很大程度上被来自相机和其他原因的噪声污染,而Max-mean和Max-median算法对这些噪声也异常敏感,所有噪声点几乎都没有滤除,从检测结果的3维图像上可以明显地看到,本文算法的检测结果非常显著,对噪声不敏感,能够实现非常好的目标检测效果,这说明本文算法与传统算法相比具有一定优势。
分别绘制这Max-mean和Max-median两种算法以及本文算法ROC曲线,如图 7所示。从前述分析可知,当ROC曲线越快趋近于顶线
4 结论
本文算法通过Facet模型提取图像方向导数信息,将红外图像方向导数特征与稀疏表示有关理论相结合,通过对单一方向上小目标特性进行详细分析,并将其扩展到多个方向上呈现的特征信息,讨论了目标和背景之间存在的区别,以稀疏表示理论为媒介,获得了小目标最终检测结果。实验结果表明,本文算法具有较高检测精度和较强抗噪声干扰能力,相比于传统检测算法具有一定优势,提高了检测率,降低了虚警率。
本文算法另一大特点在于,可根据不同情况下的检测背景,训练出不同背景字典,从而得到更好的检测效果,在实际应用中具有良好针对性。
参考文献
-
[1] Deshpande S D, Er M H, Venkateswarlu R, et al. Max-mean and max-median filters for detection of small targets[C]//Proceedings of SPIE 3809, Signal and Data Processing of Small Targets 1999. Denver, CO, USA: SPIE, 1999: 74-83.[DOI:10.1117/12.364049]
-
[2] Zeng M, Li J X, Peng Z. The design of top-hat morphological filter and application to infrared target detection[J]. Infrared Physics & Technology, 2006, 48(1): 67–76. [DOI:10.1016/j.infrared.2005.04.006]
-
[3] Gu Y F, Wang C, Liu B X, et al. A kernel-based nonparametric regression method for clutter removal in infrared small-target detection applications[J]. IEEE Geoscience and Remote Sensing Letters, 2010, 7(3): 469–473. [DOI:10.1109/LGRS.2009.2039192]
-
[4] Dong W K, Zhang J Q, Yang D D, et al. Homogeneous background prediction algorithm for detection of point target[J]. Infrared Physics & Technology, 2011, 54(2): 70–74. [DOI:10.1016/j.infrared.2010.12.037]
-
[5] Qi S X, Ming D L, Ma J, et al. Robust method for infrared small-target detection based on Boolean map visual theory[J]. Applied Optics, 2014, 53(18): 3929–3940. [DOI:10.1364/AO.53.003929]
-
[6] Zhu C, Peng Y X. Group cost-sensitive boosting for multi-resolution pedestrian detection[C]//Proceedings of the 30th AAAI Conference on Artificial Intelligence. Phoenix, Arizona: AAAI, 2016: 3676-3682. https://aaai.org/ocs/index.php/AAAI/AAAI16/paper/view/11898
-
[7] Wang X, Shen S Q, Ning C, et al. A sparse representation-based method for infrared dim target detection under sea-sky background[J]. Infrared Physics & Technology, 2015, 71: 347–355. [DOI:10.1016/j.infrared.2015.05.014]
-
[8] Wan M J, Gu G H, Qian W X, et al. Robust infrared small target detection via non-negativity constraint-based sparse representation[J]. Applied Optics, 2016, 55(27): 7604–7612. [DOI:10.1364/AO.55.007604]
-
[9] Haralick R M. Digital step edges from zero crossing of second directional derivatives[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 1984, PAMI-6(1): 58–68. [DOI:10.1109/TPAMI.1984.4767475]
-
[10] Gao C Q, Meng D Y, Yang Y, et al. Infrared patch-image model for small target detection in a single image[J]. IEEE Transactions on Image Processing, 2013, 22(12): 4996–5009. [DOI:10.1109/TIP.2013.2281420]
-
[11] Zhu C, Peng Y X. A boosted multi-task model for pedestrian detection with occlusion handling[J]. IEEE Transactions on Image Processing, 2015, 24(12): 5619–5629. [DOI:10.1109/TIP.2015.2483376]
-
[12] Bai X Z, Bi Y G. Derivative entropy-based contrast measure for infrared small-target detection[J]. IEEE Transactions on Geoscience and Remote Sensing, 2018, 56(4): 2452–2466. [DOI:10.1109/TGRS.2017.2781143]
-
[13] Costea A D, Nedevschi S. Word channel based multiscale pedestrian detection without image resizing and using only one classifier[C]//Proceedings of 2014 IEEE Conference on Computer Vision and Pattern Recognition. Columbus, OH, USA: IEEE, 2014: 2393-2400.[DOI:10.1109/CVPR.2014.307]
-
[14] Aharon M, Elad M, Bruckstein A.
$ K$ -svd:an algorithm for designing overcomplete dictionaries for sparse representation[J]. IEEE Transactions on Signal Processing, 2006, 54(11): 4311–4322. [DOI:10.1109/TSP.2006.881199] -
[15] Deng H, Sun X P, Liu M L, et al. Infrared small-target detection using multiscale gray difference weighted image entropy[J]. IEEE Transactions on Aerospace and Electronic Systems, 2016, 52(1): 60–72. [DOI:10.1109/TAES.2015.140878]
-
[16] Tropp J A, Gilbert A C. Signal recovery from random measurements via orthogonal matching pursuit[J]. IEEE Transactions on Information Theory, 2007, 53(12): 4655–4666. [DOI:10.1109/TIT.2007.909108]
-
[17] Bai X Z, Bi Y G. Derivative entropy-based contrast measure for infrared small-target detection[J]. IEEE Transactions on Geoscience and Remote Sensing, 2018, 56(4): 2452–2466. [DOI:10.1109/TGRS.2017.2781143]