-
文中基于MSTAR数据集中的SAR图像样本对所提方法进行测试。该数据集中包含了图1所示的10类车辆目标SAR图像。这些图像的分辨率达到0.3 m,直观可见目标的局部散射现象明显。文中实验中,采用所提方法对MSTAR SAR图像进行散射中心参数估计。由于这些SAR图像的散射中心参数真值未知,只能通过对比估计参数与目标局部强散射的对应性(主要是位置参数)以及重构图像与原始图像的一致性对估计结果的有效性进行初步评估。除此之外,由于属性散射中心已在SAR目标识别方法中得到广泛运用,文中通过对比文中参数估计结果与其他现有散射中心参数估计对识别结果的贡献进一步评估所提方法的有效性。
-
采用所提方法对MSTAR数据集中的一幅SAR图像进行典例验证,选取15作为散射中心数目的上限,获得具体的参数估计结果如表1所示。图2所示为位置参数在被估计SAR图像上的投影结果,可见与目标局部强散射中心吻合性较强,说明估计结果中位置参数的精度具有较高的精度。图3(a)所示为基于估计散射中心重构的目标图像,采用经典的图像相关作为测量度,计算其与被估计SAR图像的相似度达到0.92,表明重构结果与原始图像具有很强的相关性,说明散射中心各类属性参数估计结果的有效性。图3(b)为被估计图像与重构图像对比(相减)之后的残差分量。直观可见,重构残差主要是被估计图像中的背景成分,由于杂波、噪声等因素噪声。通过对被估计SAR图像中目标散射中心的估计,并据此重构,有效剔除了原始图像中存在的背景因素,对于针对性分析目标特性具有有力支撑。上述结果均表明所提方法对于SAR图像属性散射中心参数估计的有效性。
No. $ \left| A \right| $ ${x_p}/{\text{m} }$ ${y_p}/{\text{m} }$ ${L_p}/{\text{m} }$ $ \alpha $ $ {\gamma _p} $ $ {\phi _p} $ 1 513.62 −1.97 7.53 0.00 0.00 0.00 0.00 2 483.50 −8.08 12.28 2.57 0.00 0.00 0.17 3 216.39 3.18 3.75 0.00 0.00 1.82 0.00 4 131.06 10.60 −1.87 0.00 0.00 3.70 0.00 5 164.36 −5.68 6.08 0.00 0.00 0.00 0.00 6 138.23 −14.85 3.40 0.00 0.00 −0.18 0.00 7 133.72 6.99 1.63 0.00 1.00 0.00 0.00 8 523.03 −4.67 2.71 5.94 1.00 0.00 0.29 9 120.12 −13.06 7.07 0.00 0.00 0.00 0.00 10 148.23 6.69 −5.80 2.50 0.00 0.00 −0.02 11 95.01 −10.32 5.00 0.00 0.00 0.00 0.00 12 88.15 1.26 1.89 0.00 1.00 −0.19 0.00 13 86.96 0.99 7.27 0.00 1.00 −0.19 0.00 14 83.63 −2.94 10.30 0.00 1.00 0.00 0.00 15 80.11 −8.16 6.75 0.00 0.00 −0.19 0.00 Table 1. Parameter estimation results of attributed scattering centers in a MSTAR SAR image
-
参考文献[4-7]均是基于属性散射中心设计SAR目标识别方法,其中一种典型场景就是在标准操作条件下对图1所示的10类目标进行分类,具体设置的训练和测试集参见表2。为进一步验证提出的散射中心参数估计方法的有效性,可在同一个基于属性散射中心的SAR目标识别方法上采用不同估计算法的结果,进而对比识别性能。按照这一思路,文中采用现有文献中多种不同属性散射中心估计算法进行属性散射中心提取,进而采用参考文献[5]中的散射中心匹配方法对表2中的10类目标的测试样本进行分类。其中,选取的散射中心估计算法包括参考文献[8]中基于模拟退火算法、参考文献[9]中基于动态粒子群算法、参考文献[13]中的稀疏表示估计算法。在不同算法下,散射中心匹配识别方法取得的平均识别率如表3所示。对比可见,基于所提方法估计得到的属性散射中心可以取得最高的平均识别率,表明其对于目标识别的贡献更大,反映了其估计精度相对更高。因此,通过不同散射中心提取算法对于目标识别的贡献可以间接反映了所提方法具有更强的有效性。
Class BMP2 BTR70 T72 T62 BDRM2 BTR60 ZSU23/4 D7 ZIL131 2S1 Training (17°) 232 233 231 299 298 256 299 299 299 299 Testing (15°) 195 196 196 273 274 195 274 274 274 274 Table 2. Training and test samples of the ten targets
Parameter estimation algorithm Average recognition rate Simulated annealing 98.16 Dynamic particle swarm optimization 98.56 Sparse representation 98.74 Proposed 99.02 Table 3. Average recognition rates of the scattering center matching method using different parameter estimation algorithms
Application of firework algorithm into parameter estimation of attributed scattering centers in SAR images
doi: 10.3788/IRLA20210581
- Received Date: 2021-12-02
- Rev Recd Date: 2022-02-20
- Publish Date: 2022-08-31
-
Key words:
- synthetic aperture radar /
- attributed scattering center /
- parameter estimation /
- firework algorithm
Abstract: Aiming at the problem of synthetic aperture radar (SAR) attribute scattering center estimation, a method based on the firework algorithm was proposed. First, segmentation and decoupling of high-energy regions in the SAR image are performed in the image domain to obtain the representation of a single independent scattering center in the image domain. Afterwards, based on the parametric model of the attribute scattering center, an optimization problem was constructed to search for the optimal parameters of the separated single scattering center. At this stage, the firework algorithm was introduced to optimize the parameters. The algorithm has strong global and local search capabilities, and avoids falling into the local optimum thus ensuring the optimization accuracy and the reliability of the estimation of the scattering center parameters. The single scattering center after solution was eliminated from the original image, and the residual image was segmented into high-energy regions. And the attribute parameters of the next scattering center were estimated by inertia. Finally, the parameter set of all scattering centers on the input SAR image was obtained. In the implementation, the parameter estimation verification was performed based on the SAR images in the MSTAR dataset. The comparison of the parameter estimation results with the original image and the reconstruction of the original image based on the estimated parameter set reflect the effectiveness of the proposed algorithm. In addition, the experiment also validates the SAR target recognition algorithm based on the estimated attribute parameters. By comparing the recognition performance with other parameter estimation algorithms under the same condition, the performance superiority of the proposed method in the attribute scattering center parameter estimation was further demonstrated.