【作者】 周康；

【导师】 侯彪；

【作者基本信息】 西安电子科技大学 ， 电路与系统， 2018， 硕士

【摘要】 在遥感图像的超分辨率(SR)重建领域中,稀疏表示已被广泛应用于从低分辨率(LR)图像中恢复高质量的高分辨图像。由于缺少图像块之间的联系与图像内的全局信息,传统的联合字典的方法无法获得良好的SR重建结果。因此,我们提出了一种基于稀疏表示的遥感图像SR重建的有效方法。首先,我们通过LR图像的细节图像块和与之对应的HR图像块训练两个字典。其次,为了增强图像块之间的内在关系,我们引入了全局自相容模型作为全局正则化项。最后,将稀疏表示,局部约束模型和全局约束模型结合起来,提高模型的性能,并采用快速自适应收缩阈值算法(FASTA)解决GJDM中凸优化问题。与其他方法相比,该方法在保存细节和纹理信息方面表现出良好的SR重建性能,在峰值信噪比(PSNR)上有显著的提高。随着图像分辨率的显著提高,遥感图像的目标检测技术也得到了长足的发展。在很多应用场景下,通常需要更为精确的目标位置信息,而仅仅给出目标类别和粗略位置的目标检测算法还达不到这样的需求。因此,我们在Faster R-CNN基础上,通过深度卷积网络提取图像特征,候选区域,并利用分类器实现目标分类,回归器近似定位目标位置。同时在确定目标矩形框的位置后,结合目前成熟的图像分割技术,整合了目标检测与图像凸分割技术,提出了遥感图像的精致目标检测算法,一体化实现了目标分类,检测与分割。与其他目标检测算法相比较,该方法不仅能够准确的实现目标检测,同时还能提取目标的形状和轮廓特征,在精准定位与检测上达到了较高的水平。更多还原

【Abstract】 Sparse representation has been widely used in the field of remote sensing image super-resolution(SR)reconstruction to restore a high-quality image from a low-resolution(LR)image.Owing to the lack of an inner relationship between image patches and an image’s global information,the traditional methods of jointly training two over-complete dictionaries cannot obtain good SR reconstructed results.Therefore,we propose an effective approach for remote sensing image SR reconstruction based on sparse representation.First,we train two dictionaries for detail image patches and HR patches.Second,in order to enhance the inner relationship between image patches,we introduce a global self-compatibility model for global regularization.Finally,the sparse representation and local and nonlocal constraints are integrated to improve the performance of the model,and the fast adaptive shrinkage-thresholding algorithm(FASTA)is employed to solve the convex optimization problem in the GJDM.Compared to other methods,the results of the proposed method show good SR reconstruction performance in preserving details and texture information and significant improvement in peak signal to noise ratio(PSNR).With the significant improvement of remote sensing image resolution,the object detection technology has been developed.In many applications,we usually need more accurate object location information,but some object detection algorithms that only give the object category and rough location don’t meet this requirement.Therefore,on the basis of Faster R-CNN,we extract image features,region proposals by deep convolutional network,use classifier to achieve object classification,and get the object approximate position by the regression.At the same time,combining the mature image segmentation technology,we integrate the object detection and convex image segmentation technology.Then we propose the exquisite object detection algorithm of remote sensing images,and achieve the object classification,detection and segmentation integrally.Compared with other object detection algorithm,this method not only can achieve the accurate object detection,but also can extract the shape and profile characteristics of object,make the precise positioning and detection achieve a better level.更多还原