一种基于深度学习的遥感图像目标检测算法

遥感图像分析在国土资源管理、海洋监测等领域有着极为广阔的应用前景。深度学习技术已在图像处理领域取得突破性进展,然而,遥感图像固有的尺寸大、目标小而密集等特点,使得将面向普通图像的深度学习方法用于遥感目标检测普遍存在定位不准确、小目标检测难、大图检测精度差等问题。针对上述难题, 提出了一种新型遥感图像目标检测算法DFS。与传统机器学习方法相比,DFS 设计了新的维度聚类模块、定制损失函数和滑动窗口分割检测机制。其中,维度聚类模块通过设计聚类机制优化定制先验框,提高定位精度;定制损失函数提高对船只等小目标的检测精度;滑动窗口分割检测解决大图检测精度低的问题。在经典遥感数据集上开展的实验对比表明,与YOLOv2相比,DFS算法的mAP提高了256%,小目标检测效率及大图检测效能大幅提高。     

Fuzzy clustering algorithms for unsupervised change detection in remote sensing images

In this paper, we propose a context-sensitive technique for unsupervised change detection in multitemporal remote sensing images. The technique is based on fuzzy clustering approach and takes care of spatial correlation between neighboring pixels of the difference image produced by comparing two images acquired on the same geographical area at different times. Since the ranges of pixel values of the difference image belonging to the two clusters (changed and unchanged) generally have overlap, fuzzy clustering techniques seem to be an appropriate and realistic choice to identify them (as we already know from pattern recognition literatures that fuzzy set can handle this type of situation very well). Two fuzzy clustering algorithms, namely fuzzy c-means (FCM) and Gustafson-Kessel clustering (GKC) algorithms have been used for this task in the proposed work. For clustering purpose various image features are extracted using the neighborhood information of pixels. Hybridization of FCM and GKC with two other optimization techniques, genetic algorithm (GA) and simulated annealing (SA), is made to further enhance the performance. To show the effectiveness of the proposed technique, experiments are conducted on two multispectral and multitemporal remote sensing images. A fuzzy cluster validity index (Xie-Beni) is used to quantitatively evaluate the performance. Results are compared with those of existing Markov random field (MRF) and neural network based algorithms and found to be superior. The proposed technique is less time consuming and unlike MRF does not require any a priori knowledge of distributions of changed and unchanged pixels.     

基于阴影概率模型的遥感影像阴影检测方法

阴影区像素光谱响应的不一致性使得依据阈值获取的阴影检测结果与真实情况出入较大。针对这一问题,结合不透明度和亮度两种信息设计了一个全新的阴影概率模型。在此基础上,针对邻域像素信息未能充分利用的问题,提出了一个基于多尺度马尔可夫随机场(MRF)的遥感影像检测方法。首先,用所提出的模型描述多尺度影像中像素的阴影概率;然后,使用Potts模型建模多尺度标记场,同时兼顾尺度内和尺度间的邻域像素的交互关系;最后,基于最大后验(MAP)概率准则获取最终阴影检测结果。通过与色调亮度比值方法、差分双阈值方法、主成分分析法和支持向量机分类方法的对比实验证明,所提出的方法能够提高高分辨率城区遥感影像的阴影检测精度。     

A genetic algorithm with gene rearrangement for K-means clustering

In this paper, a new clustering algorithm based on genetic algorithm (GA) with gene rearrangement (GAGR) is proposed, which in application may effectively remove the degeneracy for the purpose of a more efficient search. A new crossover operator that exploits a measure of similarity between chromosomes in a population is also presented. Adaptive probabilities of crossover and mutation are employed to prevent the convergence of the GAGR to a local optimum. Using the real-world data sets, we compare the performance of our GAGR clustering algorithm with K-means algorithm and other GA methods. An application of the GAGR clustering algorithm in unsupervised classification of multispectral remote sensing images is also provided. Experiment results demonstrate that the GAGR clustering algorithm has high performance, effectiveness and flexibility.     

Many-objective evolutionary algorithm based on adaptive weighted decomposition

Decomposition is a representative method for handling many-objective optimization problems with evolutionary algorithms. Classical decomposition scheme relies on a set of uniformly distributed reference vectors to divide the objective space into multiple subregions. This scheme often works poorly when the problem has an irregular Pareto front due to the inconsistency between the distribution of reference vectors and the shape of Pareto fronts. We propose in this paper an adaptive weighted decomposition based many-objective evolutionary algorithm to tackle complicated many-objective problems whose Pareto fronts may or may not be regular. Unlike traditional decomposition based algorithms that use a pre-defined set of reference vectors, the reference vectors in the proposed algorithm are produced from the population during the search. The experiments show that the performance of the proposed algorithm is competitive with other state-of-the-art algorithms and is less-sensitive to the irregularity of the Pareto fronts.     

基于小波域压缩感知的遥感图像超分辨算法

针对单字典表达复杂多样的图像纹理存在一定的局限性的问题,利用压缩感知和小波理论建立了一种多字典遥感图像超分辨算法。首先,对训练图像在小波域的不同频带利用K-奇异值分解（K-SVD）算法建立不同的字典;然后,利用全局限制求取高分辨率图像的初始解;最后,利用正交匹配追踪算法（OMP）对初始解在小波域进行多字典稀疏求解。实验结果表明,相比基于单字典的超分辨重建算法,结果图像的主观视觉效果有很大提高,客观评价指标的峰值信噪比（PSNR）和结构相似度（SSIM）分别提高2.8 dB以上和0.01以上。字典可一次建立重复使用,降低了运算时间。     

遥感图像变化区域的无监督压缩感知

传统的基于结构特征的遥感图像变化检测方法,易受成像稳定性的影响而误差很大。针对图像内在的稀疏性结构信息,提出基于压缩感知(CS)的遥感图像变化检测方法。通过自适应构造超完备字典将图像局部信息投影到高维空间中,实现图像的稀疏表示,并运用随机矩阵得到了数据在高维空间中的低维特征子空间。最后利用模糊C均值(FCM)聚类算法进行无监督聚类,实现遥感图像变化区域信息的重构。实验结果表明,本文方法不仅能够很好的检测出图像的轮廓变化和图像的区域变化,而且对噪声具有很好的鲁棒性。     

基于小波变换的改进DnCNN网络的遥感图像去噪

遥感图像去噪一直是遥感领域的重要难题,现有的去噪算法会使图像边缘信息模糊,导致图像中有用信息丢失,为了提高遥感图像的质量,提出了一种改进DnCNN（Denoising Convolutional Neural Network）的遥感图像去噪方法,通过小波变换将原始图像分解成不同子带,采用基于遗传算法的网络结构自动搜索方法对于不同子带搜索出不同结构和参数的DnCNN网络实现去噪,使对噪声成分的提取更加有针对性。实验采用峰值信噪比（PSNR）和结构相似性（SSIM）两项评价指标对实验结果进行量化评判,标准差为20时,较原始的DnCNN方法相比PSNR值平均提高了3.5%,图像细节清晰,能有效地保护遥感图像边缘特征和轮廓结构的完整性。     

基于局部加权拟合算法的无人机遥感影像多尺度检测技术

目前提出的无人机遥感影像多尺度检测技术平均图像灰度较差,导致检测结果清晰度较低;为了解决上述问题,基于局部加权拟合算法研究了一种新的无人机遥感影像多尺度检测技术,选用最小二乘法进行多次循环计算,确定周围区域重复率,通过抽稀处理提高数据精度;根据高斯金字塔得到n阶的影像序列,利用高斯金字塔和差分尺度划分完成遥感影像的特征提取;引入加权拟合算法,构建有效影像数据集,确定影像网络模型,从而完成合并,实现影像数据的检测;实验结果表明,基于局部加权拟合算法的无人机遥感影像多尺度检测技术能够有效提高平均图像清晰度,增强检测结果的清晰度。     

遥感影像变化检测综述

变化检测作为土地利用/土地覆盖检测的关键技术,其目的是在同一区域不同时期的遥感数据中检测出变化的部分及其类型.针对传统的变化检测方法中存在繁重的人工劳动和检测结果效果差等问题,大量基于遥感影像的变化检测方法被提出.为了深入了解基于遥感影像的变化检测技术以及进一步研究变化检测方法,通过对大量有关变化检测的研究进行整理、分...     

基于改进遗传算法遥感图像非监督分类研究

传统的非监督分类方法通过人为预先设定的类别数把像素划分到相应的类别中,但类别数事先不能精确得到,因此会增大误分率,降低分类精度。提出一种新的可变聚类数目的染色体、采用Davies-Bouldin系数作为适应度,通过对传统遗传算法的一系列改进自动进化出高分辨率遥感图像的类别数和聚类中心。同时,采用整型数据来进行染色体编码,不仅降低了计算复杂度,同时也节省了存储空间。算法已用VC实现程序设计,程序结果证明该改进算法的正确性并获得令人满意的实验结果。     

基于蚁群优化与独立特征集的遥感图像实时分类算法

为了提高遥感图像的实时分类准确率与效率,提出了一种基于蚁群优化算法与独立特征集的遥感图像集实时分类算法。首先,提取遥感图像的小波域特征与颜色特征,并且组成特征向量;然后,采用蚁群优化算法对特征空间进行优化,独立地选出每个分类的显著特征集,从而降低每个子特征空间的维度;最终,每个分类独立地训练一个极限学习机分类器,从而实现对遥感图像集的分类。基于公开的遥感图像数据集进行了仿真实验,结果显示本算法实现了较高的分类准确率,并且实现了较高的计算效率。     

基于多智能体遗传的SAR图像变化检测

针对传统进化算法在SAR图像变化检测时,容易陷入局部最优,收敛速度慢,耗时过长,为了解决这些问题,提出了一种无监督的多智能体遗传SAR图像变化检测方法。利用对数比值法对预处理后的图像构造差异影像,并对差异影像进行中值滤波处理,把它的灰度值作为输入信息,通过多智能体遗传算法搜索全局阈值,根据全局阈值得到变化检测结果。仿真结果表明,该算法与GA、ICSA相比,分类准确,收敛快速,效率更高。     

基于局部-全局特征耦合与边界引导的遥感图像建筑物变化检测

针对现有变化检测方法局部特征和全局特征难以兼顾、变化对象和背景之间分界模糊的问题,提出了一种基于局部-全局特征耦合与边界引导的遥感图像建筑物变化检测方法。该方法在编码阶段采用并行的卷积神经网络和Transformer分别提取遥感图像的局部特征与全局表示。在不同尺度下,使用局部-全局特征耦合模块融合局部特征和全局特征表示,以增强图像特征的表达能力。引入边界引导分支获取变化对象的先验边界信息,使其引导变化图突出建筑物的结构特征,促进边界精确定位。该方法在LEVIR-CD和WHU数据集上进行实验验证,其F1-score分别为91.25%和91.27%,IoU分别为83.90%和83.95%。实验结果表明,该方法在检测精度上有较大的提升,且具有良好的泛化能力。     

A multiobjective fuzzy clustering method for change detection in SAR images

On account of the presence of speckle noise, the trade-off between removing noise and preserving detail is crucial for the change detection task in Synthetic Aperture Radar (SAR) images. In this paper, we put forward a multiobjective fuzzy clustering method for change detection in SAR images. The change detection problem is modeled as a multiobjective optimization problem, and two conflicting objective functions are constructed from the perspective of preserving detail and removing noise, respectively. We optimize the two constructed objective functions simultaneously by using a multiobjective fuzzy clustering method, which updates the membership values according to the weights of the two objectives to find the optimal trade-off. The proposed method obtains a set of solutions with different trade-off relationships between the two objectives, and users can choose one or more appropriate solutions according to requirements for diverse problems. Experiments conducted on real SAR images demonstrate the superiority of the proposed method.     

A recommendation framework for remote sensing images by spatial relation analysis

In recent years, Remote Sensing Images (RS-Images) are widely recognized as an essential geospatial data due to their superior ability to offer abundant and instantaneous ground truth information. One of the active RS-Image approaches is the RS-Image recommendation from the Internet for meeting the user's queried Area-of-Interest (AOI). Although a number of studies on RS-Image ranking and recommendation have been proposed, most of them only consider the spatial distance between RS-Image and AOI. It is inappropriate since both of the RS-Image and AOI not only have the spatial information but also the cover range information. In this paper, we propose a novel framework named Location-based rs-Image Finding Engine (LIFE) to rank and recommend a series of relevant RS-Images to users according to the user-specific AOI. In LIFE, we first propose a cluster-based RS-Image index structure to efficiently maintain the large amount of RS-Images. Then, two quantitative indicators named Available Space (AS) and Image Extension (IE) are proposed to measure the Extensibility and Centrality between RS-Image and AOI, respectively. To our best knowledge, this is the first work on RS-Image recommendation that considers the issues of extensibility and centrality simultaneously. Through comprehensive experimental evaluations, the experiment result shows that both indicators have their own distinguished ranking behaviors and are able to successfully recommend meaningful RS-Image results. Besides, the experimental results show that the proposed LIFE framework outperforms the state-of-the-art approach Hausdorff in terms of Precision, Recall and Normalized Discounted Cumulative Gain (NDCG).     

基于异常区域感知的多时相高分辨率遥感图像配准

在对多时相高分辨遥感图像进行配准时,由于成像条件差异,图像间存在的地物变化与相对视差偏移两类典型异常区域会影响配准精度。针对上述配准中存在的问题,提出一种基于异常区域感知的多时相高分辨率遥感图像配准方法,包括粗匹配和精配准两个阶段。尺度不变特征变换（SIFT）算法考虑到尺度空间属性,不同尺度空间提取的特征点在图像中对应不同大小的斑块,高尺度空间提取的特征点对应图像中的大斑点,其对应地物相对稳定、不易发生变化。首先,利用SIFT算法提取高尺度空间特征点完成图像快速粗匹配;其次,利用灰度相关性度量对图像块进行相对偏移量统计分类以感知视差偏移区域,同时结合空间约束条件,确定低尺度空间特征点的有效提取区域以及匹配点搜索范围,完成图像精配准。实验结果表明,将该方法用于多时相高分辨遥感图像配准,可有效抑制异常区域对特征点提取的影响进而提高配准精度。     

基于GA-EM算法的GMM遥感影像变化检测方法*

提出了一种基于GA-EM算法的高斯混合模型(GMM)遥感影像变化检测方法。该方法采用主成分分析(PCA)与传统差值法相结合的方式构造差异影像;然后使用N个成分的GMM对差异影像分布进行建模;再利用进化的迭代方法对模型进行自适应参数估计;最后利用贝叶斯准则实现变化和未变化像元分布的变化检测结果。仿真结果表明,该方法对变化目标的检测有效而可靠,具有较大的实用价值。