基于模糊粒子群优化和目标分解的极化SAR影像地物分类

发布时间：2018-04-23 19:15

本文选题：极化SAR + 分类　；参考：《西安电子科技大学》2014年硕士论文

【摘要】：极化合成孔径雷达(Polarimetric Synthetic Aperture Radar,PolSAR)因其在图像获取方面不会受到光照、天气等原因的影响,真正达到了全天侯二十四小时不间断获取图像的要求,使得极化SAR被广泛的应用到军用和民用等方面。随着获取的极化SAR影像数量的增多,同时极化SAR影像本身又包含了极为丰富的信息,因此极化SAR影像处理和解译已成为与国防建设和经济发展密切相关的一项重要研究。本文针对极化SAR影像分类方法进行了大量研究,并提出了三种改进方法:1.提出了一种基于模糊粒子群优化和散射分解的极化SAR影像地物分类的改进方法。该方法中我们采用Cloude分解对极化SAR数据进行特征分解,将得到的散射特征——散射熵和散射角作为分类依据,并采用FCM聚类算法代替复Wishart分类器对极化SAR数据进行分类,使计算复杂度降低,程序运行时间缩短。同时针对FCM聚类算法的缺陷,用粒子群算法的粒子寻优过程代替FCM中聚类中心迭代优化过程,从而避免因采用梯度下降产生的使聚类中心陷入局部最优的缺陷,最终达到比较好的分类结果。试验证明,改进后的方法得到的分类结果更加精确。2.提出了一种基于模糊粒子群优化和目标分解的极化SAR影像地物分类方法,我们继承了方法一中利用FCM聚类算法代替复Wishart分类器对极化SAR数据进行分类的方法,并且用粒子群算法的寻优迭代方法代替FCM的聚类中心更新的迭代方法。延续了方法一中优越性的同时,我们利用Cloude分解和Freeman分解,提取出极化SAR数据的散射特征——散射熵和散射功率,并将其作为分类依据。我们既可以通过Cloude分解充分利用极化SAR数据的八分量分解旋转不变性,又能通过Freeman分解合理使用幅值信息,从而保留更多图像细节,较好地保留散射机制。更加合理有效的使用极化SAR数据的散射特征使分类结果更加接近实际情况。3.提出了一种基于模糊量子粒子群优化和目标分解的极化SAR影像地物分类方法,本方法是对方法一和方法二的改进,我们在前两种方法的基础上采用改进过的粒子群算法,即量子粒子群的迭代优化过程代替FCM中聚类中心迭代优化的过程。通过将量子运动理论引入粒子群算法,利用量子运动中的粒子随机性比较大的优点,避免完全随机寻优的退化现象,基本克服了粒子群算法中过早收敛的问题。最后进行实验对比表明改进后的方法得到的分类结果更加精确。
[Abstract]:Polarimetric Synthetic Aperture Radarr PolSAR (PolSAR), because it is not affected by illumination, weather and other reasons in image acquisition, has really achieved the requirement of obtaining images continuously for 24 hours in the whole day. Polarization SAR is widely used in military and civil fields. With the increase of the number of polarimetric SAR images and the rich information contained in the polarimetric SAR images, the processing and interpretation of polarized SAR images has become an important research closely related to national defense construction and economic development. In this paper, we have done a lot of research on polarimetric SAR image classification method, and proposed three improved methods: 1. An improved method for ground object classification of polarized SAR images based on fuzzy particle swarm optimization and scattering decomposition is proposed. In this method, we use the Cloude decomposition to decompose the polarimetric SAR data, take the scattering feature-scattering entropy and the scattering angle as the classification basis, and use the FCM clustering algorithm instead of the complex Wishart classifier to classify the polarized SAR data. The computational complexity is reduced and the running time of the program is shortened. At the same time, aiming at the defects of FCM clustering algorithm, the particle optimization process of particle swarm optimization algorithm is used to replace the iterative optimization process of clustering center in FCM, so as to avoid the defect that the cluster center falls into local optimum by using gradient descent. Finally, a better classification result is achieved. The experimental results show that the classification results obtained by the improved method are more accurate. 2. 2. In this paper, a method of ground object classification for polarized SAR images based on fuzzy particle swarm optimization and target decomposition is proposed. In the first method, we use FCM clustering algorithm instead of complex Wishart classifier to classify polarized SAR data. Particle swarm optimization (PSO) algorithm is used to replace FCM's clustering center updating iteration method. At the same time, we use Cloude decomposition and Freeman decomposition to extract the scattering characteristics of polarized SAR data-scattering entropy and scattering power, and take them as the classification basis. We can not only make full use of the rotation invariance of the 8-component decomposition of polarized SAR data by Cloude decomposition, but also use the amplitude information reasonably by Freeman decomposition, so as to retain more image details and better scattering mechanism. More reasonable and effective use of polarimetric SAR data scattering features make the classification results closer to the actual situation. 3. Based on fuzzy quantum particle swarm optimization (FQPSO) and target decomposition, a new ground object classification method for polarimetric SAR images is proposed. This method is an improvement of methods 1 and 2. We use an improved PSO algorithm on the basis of the first two methods. That is, the iterative optimization process of quantum particle swarm optimization replaces the iterative optimization process of clustering center in FCM. By introducing the theory of quantum motion into particle swarm optimization (PSO) and taking advantage of the large randomness of particle in quantum motion, the degradation of complete random optimization is avoided, and the problem of premature convergence in PSO is overcome. Finally, the experimental results show that the improved method is more accurate.
【学位授予单位】：西安电子科技大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TN957.52

【相似文献】