基于卷积神经网络模型的遥感图像分类

研究了遥感图像的分类,针对遥感图像的支持向量机(SVM)等浅层结构分类模型特征提取困难、分类精度不理想等问题,设计了一种卷积神经网络(CNN)模型,该模型包含输入层、卷积层、全连接层以及输出层,采用Soft Max分类器进行分类。选取2010年6月6日Landsat TM5富锦市遥感图像为数据源进行了分类实验,实验表明该模型采用多层卷积池化层能够有效地提取非线性、不变的地物特征,有利于图像分类和目标检测。针对所选取的影像,该模型分类精度达到94.57%,比支持向量机分类精度提高了5%,在遥感图像分类中具有更大的优势。     

基于颜色-空间的图像检索算法

在分析了现有的基于内容的图像检索方法的基础上,提出了一种基于颜色空间分布的图像检索方法。该方法采用HSI颜色模型提取代表色,提高了图像检索的速度;通过图像分块后统计各方块颜色的代表色直方图并赋予不同的权值,提高了图像检索的精度,改善了图像检索的质量。实验结果表明,排序上明显优于其它几种分块算法,可以得到满意的查询结果。     

基于多层次视觉语义特征融合的图像检索算法

目的为了解决低层特征与中层语义属性间出现的语义鸿沟,以及在将低层特征转化为语义属性的过程中易丢失信息,从而会降低检索精度等问题,设计一种多层次视觉语义特征融合的图像检索算法。方法首先分别提取图像的3种中层特征(深度卷积神经网络(DCNN)特征、Fisher向量、稀疏编码空间金字塔匹配特征(SCSPM));其次,为了对3种特征进行有效融合,定义一种基于图的半监督学习模型,将提取的3个中层特征进行融合,形成一个多层次视觉语义特征,有效结合3种不同中层特征的互补信息,提高图像特征描述,从而降低检索算法中的语义鸿沟;最后,引入具有视觉特性与语义统一的距离函数,根据提取的多层次视觉语义特征来计算查询图像和训练图像的相似度量,完成图像检索任务。结果实验结果表明,与当前检索方法对比,文中算法具有更高的检索精度与效率。结论所提算法具有良好的检索准确度,在医疗、包装商标等领域具有一定的参考价值。     

小波双三次高倍插值搜索算法的遥感图像重建

由于光学遥感成像的距离较远和CCD物理尺寸的限制,获得的图像分辨率都比较低。为了在保持原始图像信息的情况下,提高图像的空间分辨率,改善图像的峰值信噪比,在深入分析前期研究中提出的小波双三次插值搜索算法的基础上,建立了小波双三次高倍插值搜索算法。该算法可以根据需要重建出原始遥感图像的高倍高分辨率遥感图像。实验结果表明,新算法重建出原始图像的2×2倍高分辨率图像的细节信息和视觉效果最好,3×3倍和4×4倍次之,5×5倍较差,更高倍数的高分辨率重建图像不宜在实际应用中采用。     

面向遥感图像检索的级联池化自注意力研究

高分辨率遥感图像检索中,由于图像内容复杂,细节信息丰富,以致通过卷积神经网络提取的特征难以有效表达图像的显著信息。针对该问题,提出一种基于级联池化的自注意力模块,用来提高卷积神经网络的特征表达。首先,设计了级联池化自注意力模块,自注意力在建立语义依赖关系的基础上,可以学习图像关键的显著特征,级联池化是在小区域最大池化的基础上再进行均值池化,将其用于自注意力模块,能够在关注图像显著信息的同时保留图像重要的细节信息,进而增强特征的判别能力。然后,将级联池化自注意力模块嵌入到卷积神经网络中,进行特征的优化和提取。最后,为了进一步提高检索效率,采用监督核哈希对提取的特征进行降维,并将得到的低维哈希码用于遥感图像检索。在UC Merced、AID和NWPU-RESISC45数据集上的实验结果表明,本文方法能够有效提高检索性能。     

基于子空间字典偶学习的高光谱图像分类

针对高光谱高分辨率带来巨大数据量和空间分辨率引起混合像元的问题,提出了基于子空间(subspace)的字典偶学习(DPL)算法,简称DPLsub算法。DPL算法是对字典学习的改进,它通过学习得到综合字典和分析字典,在模式识别中体现了高效性,而子空间投影的方法能更好地表征噪声和高度混合的像元。将光谱和空间特征融合的方法用于分类研究试验。实验数据是两幅高光谱影像,比较了子空间字典偶学习(DPLsub)模型和其他三种分类器即最小二乘支持向量机(LS-SVM)、稀疏多分类回归(SMLR)和字典学习(DL-OMP)的分类结果。实验结果显示,DPLsub算法无论在时间上还是精度上都优于其他算法,证明了这种子空间字典偶学习方法对高光谱图像分类的可行性与高效性。     

基于多残差网络的遥感图像语义分割方法

高分辨率遥感图像含有许多较为复杂的地物信息,对其进行的语义分割存在分割精度低、分割边界模糊等问题.本文提出一种新型的多尺度语义分割网络模型,旨在提高遥感图像语义分割精度.该模型为编码—解码(Encoder-Decoder)网络结构,编码器利用残差网络对图像特征进行提取;解码器利用反卷积进行上采样;残差连接将提取到的高级...     

基于特征匹配的地图图像自动配准技术研究

本文针对地图中的特征点，提出了一种基于广义特征点的图像自动配准方法，将特征点从单纯的点拓展到特征区域。以Moravec算子结合其他特征约束条件来自动搜索广义特征点。分别对两幅图像提取广义特征点后，利用基于根均方误差和交叉相关的两级匹配算法完成同名控制点的建立。并以局部加权直线拟合方法来校正图像的几何畸变。最后建立两幅图像之间的函数映射关系，完成图像的配准。实验结果证明了该方法的有效性。该方法可用于校正近景面地图影像的几何畸变和遥感图像的局部几何畸变。     

逆向学习耦合多属性查询的图像排序/检索优化算法研究

目的提出逆向学习耦合多属性查询的图像同步排序/检索优化算法,以解决当前算法检索效率与精度不高等问题。方法引入逆向学习概念,利用复杂无损函数,设计图像检索机制,优化训练误差。考虑查询项属性相关性,将训练图像分割成多个子集,联合权重因子,构造图像排序模型。对于给定的多属性查询,文中算法可以利用查询项中隐含的单词属性完成检索。结果文中算法支持多标记查询,与当前图像排序搜索机制相比,在多属性查询条件下,文中算法具有更高的检索精度(当查全率为80%时,精度较对照组分别提高了8.3%和13.2%)与效率。结论文中算法能够支持多属性查询,能进一步增加检索精度。     

改进的密集连接网络遥感图像超分辨重建

遥感图像超分辨增加了遥感图像的细节信息,在遥感图像处理中有重要的地位。为了进一步提高遥感图像超分辨的重建效果,本文提出一种改进的密集连接网络遥感图像超分辨重建算法。首先对基于残差网络的深度超分辨算法(VDSR)进行改进,结合密集连接网络(DenseNet),将残差网络中的残差块替换成密集块,并且添加一组密集层与瓶颈层,实现DenseNet网络结构的改进,同时,修改网络激活函数为PReLU函数,网络训练采用L1损失函数。为了使网络在遥感图像上具有更好的效果,训练网络时,数据集全部采用遥感图像作为训练样本。当训练的epoch达到了大约35次时网络已经收敛。实验结果表明,与VDSR算法相比,本文改进的算法对遥感图像的效果更优,峰值信噪比(PSNR)平均增加了1.05 dB,结构相似度(SSIM)平均增加了0.042。     

海洋遥感彩色图像中的海洋数据读取技术

基于对海洋遥感彩色图像的像素点数据矩阵的理解,提出了一种基于MATLAB软件的海洋遥感数据读取技术.依据这项技术,用户可以从卫星遥感获取的各种海洋要素彩色图像中读取对应于各个像素点的海洋要素数据.在MATLAB软件中,图像是由像素点构成的,全部图像信息均以矩阵的形式存储和运算.一幅图像可能包含一个数据矩阵和一个颜色映射表矩阵.用户可使用MATLAB软件读入图像,并通过所读取图像中包含的数据矩阵和颜色映射表矩阵,将图像区域中的像素点的颜色与色标中的颜色进行比较来获得像素点所表示的海洋要素数据.以一幅东中国海海域的海表面温度(SST)彩色图像实例说明了如何应用该技术来提高遥感彩色图像信息的读取能力.     

一种基于小波子带熵的遥感图像压缩算法

提出了一种使用小波子带熵进行比特分配的遥感图像压缩算法.对遥感图像进行小波提升分解后,分析了各高频子带能量百分比及其熵的变化趋势,在此基础上提出了一种新的快速比特分配方法-使用子带熵进行比特分配.然后对各个高频子带进行均匀量化,量化后的数据采用比特平面编码.对最高比特平面只记录该比特平面中非零系数的坐标,其它比特平面采用行程编码和Huffman编码方法进行压缩.实验结果表明,纹理复杂以及相对平坦的遥感图像使用该算法压缩后都可以获得很好的重构图像质量,峰值信噪比均大于34dB,而压缩比则与图像的复杂程度有关.     

A colour image retrieval scheme based on Z-scanning technique

Abstract

This paper proposes a new colour image retrieval scheme using Z-scanning technique for content-based image retrieval (CBIR). In recent years, the CBIR is a popular research topic for image retrieval. This paper proposes a scheme which employs the Z-scanning technique to extract directional intensity features for measuring the similarity between query and database images. In the multiple channel images, each colour channel can be processed individually or combined into a grey channel Y. In order to extract the features by Z-scanning technique from all images, each channel of all images must be divided into several N×N blocks. In each block, F pairs of pixels are scanned by a ‘Z’ direction to obtain the texture features. Each colour channel can be obtained an M×M Z-scanning co-occurrence matrix (ZSCM) for storing the probability of each relationship of all closest blocks. At the similarity measure stage, the ZSCMs of query image and database images are compared to measure their similarity. The experimental results show that the proposed scheme is beneficial for image retrieval when the images include the same texture or object. On the other hand, the proposed scheme also can get better retrieval results and more efficiency than colour correlogram (CC) technique for colour texture images. Another technique uses motif co-occurrence matrix (MCM) as the feature in similarity measurement. The experimental results show the proposed ZSCM can get better retrieval results and higher recall and precision values than the CC and MCM techniques for public image databases.     

A multiresolution‐based method for the determination of the relative resolution between images: First application to remote sensing and medical images

Spatial resolution is a key parameter of all kind of images. This is of particular importance in fields as, for example, medicine or remote sensing. The nominal resolution of a positron emission tomography (PET) or nuclear magnetic resonance (NMR) scanners are directly related to the size, number, and position of the detectors in the scanner ring. Also, the nominal spatial resolution of the remote sensing satellites is a well‐known characteristic because it is directly related to the area in ground that represents a pixel in the detector. Nevertheless, in practice, the actual resolution of a medical scanner image or of an image obtained from a satellite is difficult to know precisely because it depends of many other factors. However, if we have two or more images of the same region of interest, obtained using similar or different instruments, it is possible to compare the relative resolution between them. In this paper we propose a wavelet‐decomposition‐based method for the determination of the relative resolution between two images of the same area. The method can be applied, in principle, to any kind of images. As example, we applied the method to pairs of remote sensing and medical images. In the case of remote sensing, we computed the relative resolution between SPOT‐3, LANDSAT‐5 and LANDSAT‐7 panchromatic and multispectral images taken under similar as well as under very different conditions. In the case of medical imaging, we computed the relative resolution between a pair of simultaneously obtained PET and NMR images of the same object. On the other hand, if we know the true absolute resolution of one of the images of the pair, we can compute the resolution of the other. Thus, in the last part of this paper, we describe a spatial calibrator that we have designed and constructed to help compute the absolute resolution of a single remotely sensed image, presenting an example of its use. © 2006 Wiley Periodicals, Inc. Int J Imaging Syst Technol, 15, 225–235, 2005     

利用Krawtchouk矩不变量实现基于形状的图像检索

图像在经过平移、旋转和尺度变化后是否仍具有很好的检索效果是基于形状的图像检索研究的一个难点.本文提出了一种利用Krawtchouk矩不变量实现基于形状的图像检索方法.该方法首先对图像进行灰度变换,然后提取图像的低阶矩,取16个低阶矩不变量作为图像的特征向量,并按照相似性度量输出相似图像从而实现基于形状的图像检索.文中给出了实验结果,并与基于几何矩不变量和基于Zernike矩不变量的图像检索方法进行了比较.结果表明本文的方法具有更好的检索性能,和上述两种方法相比,查全率分别提高了21.52%和7.6%,查准率则分别提高了16.25%和6.25%.     

遥感图像的云分类和云检测技术研究

为了有效减小云层遮盖对遥感图像数据利用率的影响,提出了一种基于灰度特性的算法,实现了遥感图像高效自动的云分类及云检测.该方法首先将大幅遥感图像切分成小块子图,然后统计子图灰度值的均值和方差,在此基础上将云分成无云、薄云和厚云三类,最后通过边缘检测算法,实现了对厚云影响范围的有效标记.对100幅典型水域遥感图像的实验测试结果表明：正确云分类判别率达到97％,误判率小于4％,漏判率小于2％,基本满足实时性需求,证明了该算法的有效性.     

OSIS: remote sensing code for estimating aerosol optical properties in urban areas from very high spatial resolution images

The achievement of new satellite or airborne remote sensing instruments enables the more precise study of cities with metric spatial resolutions. For studies such as the radiative characterization of urban features, knowledge of the atmosphere and particularly of aerosols is required to perform first an atmospheric compensation of the remote sensing images. However, to our knowledge, no efficient aerosol characterization technique adapted both to urban areas and to very high spatial resolution images has yet been developed. The goal of this paper is so to present a new code to characterize aerosol optical properties, OSIS, adapted to urban remote sensing images of metric spatial resolution acquired in the visible and near-IR spectral domains. First, a new aerosol characterization method based on the observation of shadow/sun transitions is presented, offering the advantage to avoid the assessment of target reflectances. Its principle and the modeling of the signal used to solve the retrieval equation are then detailed. Finally, a sensitivity study of OSIS from synthetic images simulated by the radiative transfer code AMARTIS v2 is also presented. This study has shown an intrinsic precision of this tool of Δτ(a)=0.1.τ(a) ± (0.02 + 0.4.τ(a)) for retrieval of aerosol optical thicknesses. This study shows that OSIS is a powerful tool for aerosol characterization that has a precision similar to satellite products for the aerosol optical thicknesses retrieval and that can be applied to every very high spatial resolution instrument, multispectral or hyperspectral, airborne or satellite.     

Multi-Index Image Retrieval Hash Algorithm Based on Multi-View Feature Coding

In recent years, with the massive growth of image data, how to match the image required by users quickly and efficiently becomes a challenge. Compared with single-view feature, multi-view feature is more accurate to describe image information. The advantages of hash method in reducing data storage and improving efficiency also make us study how to effectively apply to large-scale image retrieval. In this paper, a hash algorithm of multi-index image retrieval based on multi-view feature coding is proposed. By learning the data correlation between different views, this algorithm uses multi-view data with deeper level image semantics to achieve better retrieval results. This algorithm uses a quantitative hash method to generate binary sequences, and uses the hash code generated by the association features to construct database inverted index files, so as to reduce the memory burden and promote the efficient matching. In order to reduce the matching error of hash code and ensure the retrieval accuracy, this algorithm uses inverted multi-index structure instead of single-index structure. Compared with other advanced image retrieval method, this method has better retrieval performance.     

空间分辨率之比对遥感图像融合质量的影响

本文主要探讨光学遥感图像融合中空间分辨率之比对融合质量的影响.采用IKONOS-2全色与多光谱图像,通过重采样的方法模拟空间分辨率之比连续变化的融合输入数据,并进行Gram-Schmidt融合实验,补充已有成果中空间分辨率之比变化不连续、融合方法单一的现状.结果表明:当空间分辨率之比降低时,融合质量随着下降,实际应用中,多光谱图像的空间分辨率越高越好;当空间分辨率之比很小时,应适当降低全色图像的空间分辨率,以减弱融合图像的光谱变形,提高融合质量;此外,即使空间分辨率之比很小,融合后图像也比融合输入多光谱图像的清晰度高,更利于图像判断与后续处理.