Print

发布时间: 2019-07-16
摘要点击次数:
全文下载次数:
DOI: 10.11834/jig.180503
2019 | Volume 24 | Number 7




    遥感图像处理    




  <<上一篇 




  下一篇>> 





区域筛选与多级特征判别相结合的PolSAR图像飞机目标检测
expand article info 韩萍, 宋厅华
中国民航大学智能信号与图像处理天津市重点实验室, 天津 300300

摘要

目的 针对全极化、复杂场景下飞机目标检测问题,提出了区域筛选与多级特征判别相结合的PolSAR飞机目标检测方法。方法 首先对原始PolSAR图像进行滤波及去取向预处理,消除相干斑和随机取向对检测效果的影响;其次对图像进行基于功率值的区域分割,提取感兴趣区域;然后对感兴趣区域进行区域筛选,提取疑似飞机目标;最后以功率交叉熵、背景匀质性、功率差异度为特征对疑似飞机目标进行筛选,得到最终的检测结果。结果 利用美国NASA实验室的AIRSAR和UVASAR系统采集的Half-Moon-Bay、Kahului及Kona地区的实测数据进行实验,并与其他方法进行了对比。在实验1中,本文方法和对比方法均能准确检测出场景中存在的2架飞机目标,本文方法产生了7个虚警,对比方法产生了22个虚警;在实验2中,本文方法和对比方法都检测出了4架飞机目标,本文方法产生了4个虚警,对比方法产生了17个虚警;在实验3中,本文方法检测出了15架飞机中的13架,产生了6个虚警,对比方法检测出了6个待测目标,产生了17个虚警。结论 本文方法在提取出疑似飞机目标的前提下,利用多种特征对疑似飞机目标进行筛选,不需要提取出机场跑道和停机坪区域,避免了由于跑道和停机坪区域提取不完整导致的检测不准确的问题,相比于对比方法,本文方法在降低虚警和漏警的同时,提高了运算效率。

关键词

PolSAR图像; 飞机目标检测; 区域筛选; 极化交叉熵; 匀质性; 功率差异度

Aircraft target detection of PolSAR image combined with regional screening and multi-feature discriminant
expand article info Han Ping, Song Tinghua
Tianjin Key Laboratory for Advanced Signal Processing, Civil Aviation University of China, Tianjin 300300, China
Supported by: National Natural Science Foundation of China (61571442)

Abstract

Objective Few studies on the target detection of PolSAR images that can be referred worldwide are currently available. In the full polarization SAR image of complex scenes, difficulties in aircraft target detection exhibit the following aspects: On the one hand, the background part includes not only the airport but also several areas, such as city, forest, mountain, ocean, and road, due to the different statistical characteristics of each area. Hence, fitting all the statistical characteristics of the background with a statistical feature is impossible. On the other hand, the polarization characteristics of the vehicle, ship, and some small buildings are particularly similar to those of the aircraft target in the SAR image. Therefore, aircraft targets are difficult to distinguish from other targets with one feature. Moreover, the shape of the aircraft target cannot be presented in the full polarization PolSAR image due to the resolution of the PolSAR data obtained by the imaging system, which can only be expressed by some pixel features. The existence of these problems complicates the detection of aircraft targets in PolSAR images. Result shows that the aircraft target exhibits several characteristics in the PolSAR image: 1) the scattering power of the aircraft target is generally higher than that of the surrounding background; 2) the aircraft target is often parked in fixed areas, such as airports and aprons, and the regions are characterized by homogeneity; and 3) the aircraft target is presented in the form of pixel blocks in the PolSAR image. Method An aircraft target detection algorithm combined with regional screening and multi-feature discriminant is proposed to solve the abovementioned problems and combine prior data. First, image preprocessing is performed to minimize the effect on the original PolSAR image due to the speckle and random orientation from target reflection. Second, the regions of interest of the runway, tarmac, and regions whose scattering properties are similar are extracted in accordance with the image power value. Then, suspected aircrafts are extracted by the area of connected domain. Finally, prior knowledge indicated that the power of the aircraft target is relatively large, the scattering power of the airport area is comparatively small, the aircraft target tail and wing roots demonstrate dihedral structural features, and the aircraft targets often appear in the airport or apron area. Thus, the suspected aircrafts are screened in terms of different characteristics, such as power cross entropy, background homogeneity, and power difference. Result Experiments were performed with the polarimetric synthetic aperture radar data from Half-Moon-Bay and Kahului Kona acquired by AIRSAR and UVASAR systems from NASA Laboratories in the United States. Few documents on the target detection of PolSAR image aircraft are available. Thus, the experiment was only compared with one method. The detection result of Half-Moon-Bay shows that both methods can accurately detect the aircraft targets. However, the proposed method produces seven false alarms. By contrast, the comparison method produces 22 false alarms. The Kahului result shows that the proposed and comparison methods can detect four aircraft targets. Nonetheless, the proposed method produces four false alarms. Contrarily, the comparison method produces 17 false alarms. The Kona result shows that the proposed method detects 13 out of 15 aircrafts and produces six false alarms. By contrast, the comparison method detects six out of 15 targets and produces 17 false alarms. The time spent in the experiment implies that the algorithm exhibits high computational efficiency. Conclusion The method eliminates the false targets by fusing different features to extract the suspected aircraft target and then obtain the final test results. The algorithm does not need to extract the airport runway and apron area, thereby avoiding inaccurate detection caused by the incomplete extraction of the apron area. The final test results produce few leak alarms, and some false alarms were generated, but the proposed method simultaneously produces fewer false and leak alarms than the comparison method. This method presents great improvement in operational efficiency because it only needs to traverse the extracted suspected target and not all the pixel points. However, the proposed algorithm still needs improvement. For example, the algorithm must be improved in terms of controlling false alarms. The false alarms generated in the algorithm are small buildings, vehicles, and ships because their characteristics in the PolSAR image are similar to the aircraft target. The parameter selection remains unable to achieve complete self-adaptation because the background area contains not only the airport but also other areas, such as urban, forests, mountains, and oceans. The background's statistical properties cannot be fitted with distribution. In addition, when the two targets are close to each other, the target background area obtained by morphological expansion may include the target to be detected, which may have a certain influence on the result. Thus, these problems must be solved in future works.

Key words

PolSAR image; aircraft target detection; regional screening; polarization cross entropy; homogeneity; power difference

0 引言

极化合成孔径雷达(PolSAR)是一种主动式成像雷达,相对于传统的光学雷达而言,不受天气、光照等条件的影响,具有全天时、全天候的特点[1],与单极化成像雷达相比,散射回波包含更多的幅度、相位等极化信息,因此在很多领域得到了广泛的应用,飞机目标检测作为其应用的一个重要领域,具有重要的军事和国防意义。

现有的SAR图像飞机目标检测的研究大多都是针对单极化功率图像,主要分为有监督和无监督检测两大类。文献[2]将扩展分型方法和恒虚警检测相结合,提出了扩展分形结合B-CFAR的SAR图像飞机目标检测算法。文献[3]提出了基于知识的飞机目标检测算法。上述方法都属于无监督检测,无监督检测的优点在于不需要大量的训练样本。文献[4]将利用先验知识得到的彩色航拍图与SAR图像相结合,提出智能融合的飞机目标检测方法。文献[5]提出基于卷积神经网络的SAR图像飞机目标检测方法。这两种方法属于有监督检测。

针对PolSAR图像飞机目标检测,目前研究成果较少,文献[6]根据失事飞机目标通常尾部都保留完好这一信息,分析了尾翼在SAR图像中体现出的二面角结构这一特征,并利用该特征对失事飞机目标进行检测。加拿大实验室通过分析已有的实验数据,给出了可以用于失事飞机检测的检测方案[7];文献[8]和文献[9]同样分析了失事飞机残骸尾部的散射特征,验证了该特征有助于失事飞机检测。上述的研究对象均是失事飞机,对于正常的飞机目标检测,文献[10]提出了基于极化交叉熵和Yamaguchi分解的飞机目标检测方法,该方法无需训练样本,能够直接从PolSAR图像中检测出飞机目标,不足之处在于,算法需要对图像进行遍历检测,运算效率相对较低,同时产生的虚警较多。

PolSAR图像飞机目标检测的难点在于:1)受PolSAR成像系统采集的数据分辨率的影响,飞机目标的几何特征无法体现,仅表现为一些像素特征;2)目前已有的检测方法是在提取出机场跑道感兴趣区域后,再对飞机目标进行检测,若这些区域提取的不理想,容易产生漏警;3)大场景下PolSAR图像飞机目标检测时,背景部分不仅包含机场,还包含城区、森林、山地、海洋等其他区域,背景的统计特性无法用一种分布去拟合。

针对上述问题,本文提出了一种区域筛选与多级特征判别相结合的飞机目标检测算法。该算法首先对PolSAR图像进行预处理,去除相干斑和地物随机取向对检测结果的影响;再对图像进行基于功率值的区域分割,将功率值较大的区域分割出来,然后进行区域筛选,提取出疑似飞机目标;最后以功率交叉熵、背景匀质性、功率差异度为特征对疑似飞机目标进行检测,得到检测结果。该方法仅对粗提取得到的疑似飞机目标进行判别,不需要对图像中所有的像素点进行遍历,在检测飞机目标的同时提高了运算效率。

1 特征提取

1.1 极化交叉熵

PolSAR图像中,目标的相干矩阵$\mathit{\boldsymbol{T}}$表示为

$ \mathit{\boldsymbol{T}} = \left[ {\begin{array}{*{20}{c}} {{T_{11}}}&{{T_{12}}}&{{T_{13}}}\\ {{T_{21}}}&{{T_{22}}}&{{T_{23}}}\\ {{T_{31}}}&{{T_{32}}}&{{T_{33}}} \end{array}} \right] $ (1)

设任意两个目标的相干矩阵分别为$\mathit{\boldsymbol{T}}_a$$\mathit{\boldsymbol{T}}_b$,则两目标之间的散射相似性参数22

$ r\left( {{\mathit{\boldsymbol{T}}_a},{\mathit{\boldsymbol{T}}_b}} \right) = \frac{{\left| {{\rm{tr}}\left( {{\mathit{\boldsymbol{T}}_a} \cdot {\mathit{\boldsymbol{T}}_b}} \right)} \right|}}{{\sqrt {{\rm{tr}}\left( {{\mathit{\boldsymbol{T}}_a} \cdot {\mathit{\boldsymbol{T}}_a}} \right){\rm{tr}}\left( {{\mathit{\boldsymbol{T}}_b} \cdot {\mathit{\boldsymbol{T}}_b}} \right)} }} $ (2)

式中,${\rm{tr}}(·)$表示矩阵的迹。

由于实际目标的结构错综复杂,导致对目标散射特征解译的准确性也变得相当困难。通常借助计算目标与标准散射体的相似性参数来分析实际目标的散射特性。机场区域中,停机坪或跑道区域的散射特性主要体现为平面散射。对于飞机目标,机翼与机身构成多个二面角结构,当目标结构中包含多个二面角结构时会产生螺旋体散射[12],通过计算与典型二面角散射、左螺旋散射、右螺旋散射的相关性,可以提取飞机目标特征。已知二面角面散射体、左螺旋体、右螺旋体的散射相干矩阵分别为

$ {\mathit{\boldsymbol{T}}_{\rm{d}}} = \left[ {\begin{array}{*{20}{c}} 0&{}&{}\\ {}&1&{}\\ {}&{}&0 \end{array}} \right] $ (3)

$ {\mathit{\boldsymbol{T}}_1} = \left[ {\begin{array}{*{20}{c}} 0&{}&{}\\ {}&1&{ - j}\\ {}&j&1 \end{array}} \right] $ (4)

$ {\mathit{\boldsymbol{T}}_{\rm{r}}} = \left[ {\begin{array}{*{20}{c}} 0&{}&{}\\ {}&1&j\\ {}&{ - j}&1 \end{array}} \right] $ (5)

根据式(2)得到任意目标的相干矩阵$\mathit{\boldsymbol{T}}$$\mathit{\boldsymbol{T}}_{\rm{d}}$$\mathit{\boldsymbol{T}}_{\rm{l}}$$\mathit{\boldsymbol{T}}_{\rm{r}}$的散射相似性为

$ {r_1} = r\left( {\mathit{\boldsymbol{T}},{\mathit{\boldsymbol{T}}_{\rm{d}}}} \right) = \frac{{{T_{22}}}}{{{T_{11}} + {T_{22}} + {T_{33}}}} $ (6)

$ {r_2} = r\left( {\mathit{\boldsymbol{T}},{\mathit{\boldsymbol{T}}_1}} \right) = \frac{{{T_{22}} + {T_{33}} - 2{\mathop{\rm Im}\nolimits} \left( {{T_{23}}} \right)}}{{{T_{11}} + {T_{22}} + {T_{33}}}} $ (7)

$ {r_3} = r\left( {\mathit{\boldsymbol{T}},{\mathit{\boldsymbol{T}}_{\rm{r}}}} \right) = \frac{{{T_{22}} + {T_{33}} + {\mathop{\rm Im}\nolimits} \left( {{T_{23}}} \right)}}{{{T_{11}} + {T_{22}} + {T_{33}}}} $ (8)

为了衡量目标散射特性与背景散射特性的差异性,引入统计量极化交叉熵($\mathit{\boldsymbol{p}}$)[13]

$ p = \sum\limits_{i = 1}^3 {{r_{t,i}}\ln \left( {\frac{{{r_{t,i}}}}{{{r_{c,i}}}}} \right)} $ (9)

式中,$r_{t, i}$$r_{c, i}$分别表示目标和背景与二面角散射体、左螺旋体、右螺旋体的散射相似性参数,目标与背景示意图如图 1所示,$p$越大,表明目标与背景的差异度越明显。

图 1 目标与背景区域示意图
Fig. 1 Diagram of the target area and background area

1.2 背景匀质性

为了表征目标背景的匀质性,引入背景匀质性($\mathit{\boldsymbol{v}}_i$)的概念[14]$\mathit{\boldsymbol{v}}_i$是由背景均值和方差构成的统计量,具体计算方式为

$ {v_i} = 1 + \frac{{{{\bar \sigma }^2}}}{{{{\bar \mu }^2}}} $ (10)

式中,$\overline{\mu}$$\overline{\sigma}^{2}$分别表示目标背景区域内像素功率的均值和方差。根据$\mathit{\boldsymbol{v}}_i$表达式可知,若$\overline{\sigma}^{2}=0$,则$\mathit{\boldsymbol{v}}_i=1$,说明背景是均匀的;若$\overline{\sigma}^{2}$越大,则$\mathit{\boldsymbol{v}}_i>1$,说明背景越不均匀。

1.3 功率差异度

$\mathit{\boldsymbol{T}}$矩阵可得目标的散射功率为

$ s = {T_{11}} + {T_{22}} + {T_{33}} $ (11)

为了描述目标功率与背景功率的差异性,引入功率差异度($P$),表示形式为

$ P = {\bar s_1} - {\bar s_2} $ (12)

式中,$\overline{s_{1}}$表示目标区域所有像素点功率的均值,$\overline{s_2}$表示目标背景区域所有像素点功率的均值,$P$的值越大,表明目标与背景的差异度越大。

2 本文算法

算法首先对PolSAR图像进行预处理,然后利用阈值分割法提取出感兴趣区域,在此基础上利用区域筛选的方法提取出疑似飞机目标,最后利用已知的先验知识对疑似飞机目标进行特征筛选,得到最终的检测结果,具体实现流程如图 2所示。

图 2 算法流程图
Fig. 2 Flowchart of the algorithm

1) 预处理。为了消除目标相干斑和随机取向对检测结果的影响,首先对图像进行滤波处理[15],然后利用去取向方法[16]解决图像的随机取向问题。

2) 提取感兴趣区域(ROI)。由于飞机蒙皮大部分由复合金属材料制成,在PolSAR图像中表现为强散射,利用阈值法提取出功率较大的感兴趣区域,即

$ {x_{\left( {i,j} \right)}} = \left\{ \begin{array}{l} 1\;\;\;\;s\left( {i,j} \right) > {t_1}\\ 0\;\;\;\;其他 \end{array} \right. $ (13)

式中,$t_1$为功率筛选的阈值,$s_{(i, j)}$为待处理像素点的归一化功率值,$x_{(i, j)}=1$表示保留该像素点,$x_{(i, j)}=0$表示去除该像素点,遍历整幅图像后即形成一幅二值图像。

3) 提取疑似飞机目标。对步骤2)得到的二值图进行连通域计算,假设第$j$个连通域中像素点的个数为$a$($j$),利用飞机目标在图像中是孤立的点目标这一特征,并结合图像分辨率及典型飞机目标实际标准尺寸,通过区域筛选的方法剔除ROI中非点目标区域,进而提取出疑似飞机目标。具体方法为

$ A\left( j \right) = \left\{ \begin{array}{l} 1\;\;\;\;a\left( j \right) \in \left[ {{a_1},{a_2}} \right]\\ 0\;\;\;\;其他 \end{array} \right. $ (14)

式中,$A(j)$=1,表示第$j$个连通域为疑似飞机目标;$A(j)$=0,表示第$j$个连通域不是待检测的飞机目标。

4) 形态学处理。目的是为了提取疑似飞机目标周围的背景区域,参考加窗原理将第1次膨胀得到的区域作为过渡区域处理,同时为了尽可能地避免邻近的飞机目标出现在背景区域中,本文对飞机目标进行两次形态学膨胀处理,将第2次膨胀得到的区域作为疑似飞机目标的背景区域。形态学膨胀的计算方法[17]

$ \mathit{\boldsymbol{Y}} = \mathit{\boldsymbol{E}} \oplus \mathit{\boldsymbol{\theta }} = \left\{ {y:\mathit{\boldsymbol{\theta }}\left( y \right) \cap \mathit{\boldsymbol{E}} \ne \emptyset } \right\} $ (15)

式中,$⊕$表示异或运算,$\mathit{\boldsymbol{\theta }}$表示结构元素,$\mathit{\boldsymbol{\theta }} = \left[ {\begin{array}{*{20}{l}} 0&1&0\\ 1&1&1\\ 0&1&0 \end{array}} \right], \mathit{\boldsymbol{E }}$表示待处理的疑似飞机目标区域的二值图,$\mathit{\boldsymbol{Y}}$表示形态学膨胀后的飞机目标区域。

5) 目标特征计算。根据步骤3)、4)得到的飞机目标与背景区域,按照第2节给出的特征计算方法,计算特征$v_i$$p$$p$

6) 特征筛选。由飞机目标先验信息可得:(1)飞机相对于其所处的区域的散射功率较大;(2)飞机停放区域具有匀质性;(3)飞机目标的极化交叉熵相对较大。可以通过上述信息对疑似飞机目标进行特征筛选,得到最终的检测结果。具体方法为

$ {L_j} = \left\{ \begin{array}{l} 1\;\;\;\;{v_{ij}} < {t_2}\;且\;{P_j} > {t_3}\;且\;{p_j} > {t_4}\\ 0\;\;\;\;其他 \end{array} \right. $ (16)

式中,$L_j$表示对第$j$个疑似飞机目标的检测标记值,$t_2$$t_3$$t_4$为特征筛选时所选用的阈值。当第$j$个疑似飞机目标的3个待检测特征满足上述关系时,则将该疑似飞机目标设为1,即表明该疑似飞机目标是要检测的飞机目标;否则设为0,表明该疑似飞机目标不是要检测的飞机目标。阈值$th2$$th3$$th4$的设置方法如下:

1) 将所有疑似飞机目标的背景匀质性($v_i$)、功率差异度($p$)、极化交叉熵($p$)等特征分别按从小到大顺序进行排序,构成一维矩阵$\mathit{\boldsymbol{V}}$$\mathit{\boldsymbol{P}}$$\mathit{\boldsymbol{P}}$,其中,$\mathit{\boldsymbol{V}} = \left[ {{v_1}, {v_2}, \cdots, {v_n}} \right], \mathit{\boldsymbol{P}} = \left[ {{P_1}, {P_2}, \cdots, {P_n}} \right], \mathit{\boldsymbol{P}} = \left[ {{p_1}, {p_2}, \cdots, {p_3}} \right]$

2) 选取$\mathit{\boldsymbol{V}}$, $\mathit{\boldsymbol{P}}$, $\mathit{\boldsymbol{P}}$序列中某处的值作为阈值$t_2$, $t_3$, $t_4$的估计值。

3 结果分析

3.1 实验数据

为了证明本文方法的有效性,对多幅实测数据图像进行了实验,本文给出两组实验结果,并与文献[10]的实验结果进行对比。

实验1:实验数据来自美国AIRSAR系统采集的Hawaii地区的16视全极化Half-Moon-Bay的SAR数据,图像大小为830×401像素,图像距离向分辨率为6.66 m,方位向分辨率为8.21 m,已知图像场景包含了2架飞机,型号分别为Beechcraft(比奇)和Cessna(赛斯纳),原始Pauli图及局部放大图如图 3(a)所示,图 3(b)为该区域对应的光学参考图(非同一时间采集),此外还有机场跑道、建筑物、植被、车辆、海洋、城区、船舶、反射器等目标。

图 3 Half-Moon-Bay地区实验数据
Fig. 3 The of experimental data Half-Moon-Bay area ((a) original Pauli image; (b) optical reference image)

实验2:实验数据来自美国UAVSAR系统采集的Hawaii地区L波段的4视全极化Kahului机场数据,图像大小为344×203像素,图像距离向分辨率为7.2 m,方位向分辨率为4.99 m,图像场景包含了4架飞机、原始Pauli图及局部放大图,如图 4(a)所示,图 4(b)为该区域对应的光学参考图(非同一时间采集),此外还有机场跑道、建筑物、植被、车辆、海洋等目标。

图 4 Kahului地区实验数据
Fig. 4 The experimental data of Kahului area ((a) original Pauli image; (b) optical reference map)

实验3:实验数据来自美国UAVSAR系统采集的L波段的4视全极化Kona国际机场数据,图像大小为412×501像素,图像距离向分辨率为7.2 m,方位向分辨率为4.99 m,图像场景包含了大约15架飞机,图 5(a)中用椭圆框进行了标注,原始Pauli图及局部放大图如图 5(a)所示,图 5(b)为该区域对应的光学参考图(非同一时间采集),此外该区域还包含城区、车辆、林地、海洋等目标。

图 5 Kona国际机场实验数据
Fig. 5 The experimental data of Kona international airport area ((a) original Pauli image; (b) optical reference image)

3.2 实验结果

目标在PolSAR图像中的散射功率强度不仅会受到入射角及散射雷达散射横截面积的影响,还会受到目标结构特征的影响。飞机目标只有翼根和尾翼这些易形成二面角处的散射功率回波在PolSAR图像中才表现为强散射区域,而这些强散射功率区域是以块状区域形式呈现的,块状区域的大小会受到雷达分辨率的影响,现役的民用飞机的翼展普遍在15~80 m之间,机身长普遍在10~75 m之间,在AIRSAR系统和UVASAR系统采集的PolSAR图像中,大多数飞机目标呈现为强散射功率块状区域内的像素点个数在3~25之间,因此本文在区域筛选时,选用的阈值$a_1=3$$a_2=25$

图 6为实验1中Half-Moon-Bay地区的实验结果,th1 = 0.3,th2,th3,th4的估计值分别在$\mathit{\boldsymbol{V}}$$\mathit{\boldsymbol{P}}$$\mathit{\boldsymbol{P}}$序列的2/3,1/3,3/4处取得,图 6(a)为该地区的原始功率图,图 6(b)为功率筛选后的感兴趣区域二值图,图 6(c)为区域筛选后的二值图,最终检测结果如图 6(d)所示,图 6(e)给出了文献[10]方法的检测结果。其中标号1,2为真实的飞机目标,矩形框中为产生的虚警。

图 7为实验2中Kahului地区的实验结果,$th1 = 0.3, {\rm{ }}th2, {\rm{ }}th3, {\rm{ }}th4$的估计值分别在$\mathit{\boldsymbol{V}}$$\mathit{\boldsymbol{P}}$$\mathit{\boldsymbol{P}}$序列的2/3,3/4,3/4处取得。图 7(a)为该地区的原始功率图,图 7(b)为功率筛选后的感兴趣区域二值图,图 7(c)为区域筛选后的二值图,最终检测结果如图 7(d)所示,图 7(e)给出了文献[10]方法的检测结果。其中标号1,2,3,4为真实的飞机目标,矩形框中为产生的虚警。

图 6 Half-Moon-Bay地区实验结果
Fig. 6 The experiment results of Half-Moon-Bay ((a) original power map; (b) binary map of region of interest after power screening; (c) binary image of suspected aircraft target after regional screening; (d) ours; (e) detection results of reference [10])
图 7 Kahului地区实验结果
Fig. 7 The experiment results of Kahului ((a) original power map; (b) binary map of region of interest after power screening; (c) binary image of suspected aircraft target after regional screening; (d) ours; (e) detection results of reference [10])

图 8为实验3中Kona国际机场数据的实验结果,$t_{1}=0.02, t_{2}, t_{3}, t_{4}$的估计值分别在$\mathit{\boldsymbol{V}}$$\mathit{\boldsymbol{P}}$$\mathit{\boldsymbol{P}}$序列的2/3,1/2,1/3处取得。图 8(a)为该地区的原始功率图,图 8(b)为功率筛选后的感兴趣区域二值图,图 8(c)为区域筛选后的二值图,最终检测结果如图 8(d)所示,图 8(e)给出了文献[10]方法的检测结果。椭圆框中为检测出的飞机目标,方框中为产生的虚警。

图 8 Kona国际机场地区实验结果
Fig. 8 The experiment results of Kona international airport area ((a) original power map; (b) binary map of region of interest after power screening; (c) binary image of Suspected aircraft target after regional screening; (d) ours; (e) detection results of reference [10])

3.3 实验分析

从实验1和实验2的检测结果可以看出,本文方法和文献[10]方法均能准确检测出待检测的飞机目标,但本文的实验结果产生的虚警相对较少,在实验3的检测结果中,本文方法产生的漏警、虚警都相对较少。本文方法将目标以区域整体进行处理,降低了车辆、小型建筑物的影响,减少了虚警率,此外本文算法仅对疑似飞机目标进行处理、检测,无需对图像进行遍历,降低了算法的复杂度,提高了运算效率。表 1给出了两种方法针对上述3组实验的具体数据对比。

表 1 检测结果对比
Table 1 Comparative of detection results

下载CSV
方法 飞机目标 检测结果 虚警 漏警 耗时/s
实验1 本文 2 2 7 0 2.12
文献[10] 2 2 22 0 16.96
实验2 本文 4 4 4 0 7.29
文献[10] 4 4 17 0 77.01
实验3 本文 15 13 6 2 7.14
文献[10] 15 6 17 9 44.12

4 结论

针对全极化、复杂场景下的SAR图像飞机目标检测问题,提出一种区域筛选与多级特征判别相结合的飞机目标检测算法。算法利用区域筛选的方式筛选出疑似飞机目标,然后利用PolSAR数据中飞机目标特有的特征对疑似飞机目标进行筛选,得到最终的检测结果,最后利用实测数据进行实验。从实验结果的对比分析中可以得知,本文方法的最终检测结果与对比方法相比,在漏警、虚警的抑制和运算效率提升方面都有很大改善。但是本文算法在虚警的控制方面还有待提升,算法中产生的虚警是一些小型建筑物、车辆、舰船,这是由于它们在PolSAR图像中表现出的特性与飞机目标较为相似。此外,算法在阈值参数选择方面还不能实现完全自适应,这是由于背景区域不仅包含机场,还包含城区、森林、山地、海洋等其他区域,其统计特性无法用一种分布去拟合,因此如何更好地区分飞机和舰船、车辆、建筑物之间差别,自适应地选取阈值将是下一步工作的重点。

参考文献

  • [1] Lee J S, Pottier E. Polarimetric Radar Imaging:from Basics to Applications[M]. Boca Raton: CRC Press, 2009: 4-6.
  • [2] Hu X B, Wu M Q, Zhang C Y. Application of extended fractal with B-CFAR to target detection on SAR image[J]. Radar Science and Technology, 2004, 2(5): 279–283. [胡笑斌, 吴曼青, 张长耀. 扩展分形法结合B-CFAR在SAR图像目标检测中的应用[J]. 雷达科学与技术, 2004, 2(5): 279–283. ] [DOI:10.3969/j.issn.1672-2337.2004.05.006]
  • [3] Wang W G, Sun J P, Mao S Y. Knowledge-based plane detection from SAR image[J]. Signal Processing, 2009, 25(8A): 623–626. [王文光, 孙进平, 毛士艺. 基于知识的SAR图像飞机检测[J]. 信号处理, 2009, 25(8A): 623–626. ]
  • [4] Filippidis A, Jain L C, Martin N. Fusion of intelligent agents for the detection of aircraft in SAR images[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2000, 22(4): 378–384. [DOI:10.1109/34.845380]
  • [5] Wang S Y, Gao X, Sun H, et al. An aircraft detection method based on convolutional neural networks in high-resolution SAR images[J]. Journal of Radars, 2017, 6(2): 195–203. [王思雨, 高鑫, 孙皓, 等. 基于卷积神经网络的高分辨率SAR图像飞机目标检测方法[J]. 雷达学报, 2017, 6(2): 195–203. ] [DOI:10.12000/JR17009]
  • [6] Sadjadi F A. Detection of aircraft crash sites from space using fully polarimetric SIR-C SAR imagery for search and rescue applications[C]//Proceedings Volume 3718, Automatic Target Recognition IX. Orlando, FL: SPIE, 1999: 213-220.[DOI:10.1117/12.359953]
  • [7] Lukowski T I, Charbonneau F J. Synthetic aperture radar and search and rescue:detection of crashed aircraft using imagery and interferometric methods[J]. Canadian Journal of Remote Sensing, 2002, 28(6): 770–781. [DOI:10.5589/m02-070]
  • [8] Han P, Ge P, Wu R B. Crashed plane detection based on the fusion of CFAR detection and polarimetric decomposition[C]//Proceedings of 2011 IEEE CIE International Conference on Radar. Chengdu, China: IEEE, 2011: 744-747.[DOI:10.1109/CIE-Radar.2011.6159648]
  • [9] Han P, Zhang X S, Ge P. Crashed airplane detection based on feature fusion in POLSAR image[C]//Proceedings of the 11th International Conference on Signal Processing. Beijing, China: IEEE, 2012: 2003-2006.[DOI:10.1109/ICoSP.2012.6491973]
  • [10] Han P, Chen L L. Airplane detection algorithm of polarimetric SAR image based on polarimetric cross entropy and Yamaguchi decomposition[J]. Signal Processing, 2017, 33(3): 389–396. [韩萍, 陈丽丽. 基于极化交叉熵和Yamaguchi分解的飞机目标检测方法[J]. 信号处理, 2017, 33(3): 389–396. ] [DOI:10.16798/j.issn.1003-0530.2017.03.020]
  • [11] Yang J, Peng Y N, Lin S M. Similarity between two scattering matrices[J]. Electronics Letters, 2001, 37(3): 193–194. [DOI:10.1049/el:20010104]
  • [12] Wang C, Zhang H, Chen X, et al. Fully Polarized Synthetic Aperture Radar:Image Processing[M]. Beijing China: Science Press, 2008: 75-76. [ 王超, 张红, 陈曦, 等. 全极化合成孔径雷达图像处理[M]. 北京: 科学出版社, 2008: 75-76.]
  • [13] Xu J Y, Yang J, Peng Y N, et al. Using cross-entropy for polarimetric radar discrimination problem[J]. Electronics Letters, 2002, 38(12): 593–594. [DOI:10.1049/el:20020383]
  • [14] Xu C A, He Y, Yao Q Q, et al. A modified robust CFAR detector[C]//Proceedings of 2011 International Conference on Electronic & Mechanical Engineering and Information Technology. Harbin, China: IEEE, 2011: 4439-4442.[DOI:10.1109/EMEIT.2011.6023129]
  • [15] Han P, Han B B. Lee filter of PolSAR image based on typical scattering difference index[J]. Systems Engineering and Electronics, 2018, 40(2): 287–294. [韩萍, 韩宾宾. 基于典型散射差异指数的PolSAR图像Lee滤波[J]. 系统工程与电子技术, 2018, 40(2): 287–294. ] [DOI:10.3969/j.issn.1001-506X.2018.02.08]
  • [16] An W T, Cui Y, Yang J. Three-component model-based decomposition for polarimetric SAR data[J]. IEEE Transactions on Geoscience and Remote Sensing, 2010, 48(6): 2732–2739. [DOI:10.1109/TGRS.2010.2041242]
  • [17] Cheng K, Cheng Y. Fast parallel algorithm of morphology operation based on element subsection[J]. Video Engineering, 2016, 40(3): 26–29. [程楷, 程勇. 基于结构元素分解的二值形态学并行快速算法[J]. 电视技术, 2016, 40(3): 26–29. ] [DOI:10.16280/j.vide0e.2016.03.006]