An effective network structure is essential for the classification of satellite image time series (SITS). Deep learning models have been widely used for SITS classification and achieved impressive performance, especially the architectures based on self-attention. However, the lack of efficient and comprehensive attention to valuable bands and time series structure hinders the performance to some extent. To address this problem, an end-to-end attention-aware dynamic self-aggregation network (ADSN) is proposed for SITS classification in this work, which combines two main parts: spectral focusing and spectral–temporal feature learning. The core components of ADSN are the channel attention module and dynamic self-aggregation block. Specifically, informative bands in the SITS flowing through the channel attention module can adaptively get a high weight to increase their contributions, while the attentions of some low-efficiency bands are weakened. Besides, the dynamic self-aggregation block, which integrates multiscale dynamic convolution and improved multihead attention in parallel, can simultaneously capture long- and short-distance sequence structures and position relationships to better represent temporal information. Compared with random forest (RF) and seven deep learning algorithms, the proposed model effectively learns spectral and temporal features, and the experimental results confirm that ADSN has achieved superior classification accuracy and generalization ability on two SITS datasets with extremely unbalanced samples.

合成孔径雷达（SAR）因其对地观测全天候、全天时优势,成为多云多雨天气限制下洪水动态监测中不可或缺的数据来源之一。由于GEE （Google Earth Engine）云计算平台的兴起和短重访Sentinel-1数据的可获取性,洪水监测与灾害评估目前正面向动态化、广域化快速发展。顾及洪水淹没区土地覆盖变化的复杂性和发生时间的不确定性,基于时序Sentinel-1A卫星数据提出了针对大尺度范围、连续长期的汛情自动检测及动态监测方法。该方法首先,利用图像二值化分割时序SAR数据实现水体时空分布粗制图,逐像素计算时间序列中被识别为水体候选点的频率。然后,利用Sentinel-2光学影像对精度较粗的初期SAR水体提取结果进行校正,得到精细的水体分布图。最后,针对不同频率区间的淹没特点,采用差异化的时序异常检测策略识别淹没范围:对低频覆水区利用欧氏距离检测时序断点,以提取扰动强度大、淹没时间短的洪涝灾害区;对高频覆水区利用标准分数（Z-Score）检测时序断点,以提取季节性水体覆盖区。在GEE平台上利用该方法,实现了2020-05—10长江中下游地区全域洪水淹没范围时空信息的自动、快速、有效监测,揭示了不同区域汛情发展模式的差异性。本文提出的洪水快速监测方法对大尺度下的汛情动态监测、灾害定量评估和快速预警响应具有重要的现实意义。

The glacier environment in the Namcha Barwa-Gyala Peri (NBGP) massif is regarded as one of the most sensitive areas to climate change, yet the change estimates remain inadequate due to the limited knowledge of the complex debris-covered glacier environment and the lack of available remote sensing data caused by continuously cloudy weather. To examine the changes of the complex glacial environment, a multiple hierarchical object- and rule-based classification (MHORC) scheme is proposed by combining Landsat time series images and topographic data. Object-based image analysis (OBIA) is introduced due to its capabilities of handling data contextually and hierarchically and using multi-source data jointly so that multiple hierarchical rule-based classification can be developed to distinguish the objects with similar spectrum such as debris-covered glaciers and their surrounding terrain. All the available Landsat images from 1987 to 2019 are filtered by the cloud cover condition (less than 10%) and accordingly used in the MHORC to obtain spatial and temporal information of the glacier and snow over the NBGP massif. The images collected in the snow-free season are used to estimate the glacier extent state and the historical changes, and those in the snow-covered are used to estimate the snow cover fluctuation using harmonic analysis. The results show that the NBGP contains 65 glaciers(>= 0.1 km(2)) with a total area of 462.7 km(2), among which valley glaciers account for 21.5% in number but cover an area of 354.2 km(2), more than 76% of the total area. An accelerating reduction trend of the glacier area is confirmed as the area change velocity is -0.62 km(2) yr(-1) between 1999 and 2003 and reach -2.95 km(2) yr(-1) between 2003 and 2015. The seasonal snow cover area shows a reduction in the secular trend and an increase in fluctuation amplitude. Both glacier and seasonal snow display heterogeneity in the spatial and temporal patterns. Such heterogeneous dynamics might be related to the trade-off between water vapor supply and regional thermal conditions. This MHORC developed in this study demonstrates its effectiveness in delineating the debris-covered glacier environment and the potential of using Landsat time series to track glacial environment evolution. Besides, the analysis of seasonal snow fluctuations provides a novel perspective to re-recognize the glacial environment dynamics, and the knowledge of glacier extent changes and season snow fluctuations over the NBGP massif can improve the understanding of environmental effects from climate change.

常态化地理国情监测能够全面、动态地掌握地理国情信息及其变化,为经济建设和社会发展提供数据基础.地理国情普查成果是按照统一规范标准、经过内业解译和外业核查形成的矢量数据.如何在普查成果的基础上,利用多时相遥感影像实现变化信息提取与更新是地理国情监测的关键.针对地理国情普查成果的特点与监测需求,以多时相遥感影像处理分析为基础,构建了针对地理国情监测的变化检测方法体系,提出了像元—对象结合的多时相影像变化检测、基于对象实体统计分析的变化识别方法,实现了综合地理国情普查成果和遥感影像的地理国情变化检测与数据更新.基于像元—对象结合的多时相影像变化检测首先根据传统的变化矢量分析法提取基于像元的变化检测结果,再以地理国情普查的矢量对象为统计单元计算对象内变化像元的比例,以此判断该矢量对象是否发生了变化,并根据变化像元的比例计算其变化强度.基于对象实体统计分析的变化检测方法直接以地理国情矢量为对象进行特征提取和差异构造,再将差异影像进行阈值分割得到基于地理国情对象的变化检测图.最后,根据变化检测结果,对变化区域进行面向对象分割,并从上一期未变化区域选取训练样本训练分类器模型以得到变化区域的地表覆盖类型,将变化区域与未变化区域结合得到更新后的地理国情矢量图.选取江阴市地理国情普查成果和两期高分辨率遥感影像进行试验,结果表明本文提出的方法在准确提取和解释变化区域的同时,明显提高了变化检测和数据更新的效率,可用于常态化地理国情监测.

Change vector analysis (CVA) is an effective and widely used unsupervised change detection algorithm in remote sensing. It separates changed pixels from unchanged pixels by binarizing bi-temporal difference image. However, the results and performance are affected by the image acquisitions at different dates and the threshold decision rules for change magnitudes, resulting in serious false and missed detections. This paper proposed a novel tri-temporal logic-verified change vector analysis (TLCVA) approach which can identify the errors of CVA through logical reasoning and judgement with an additional temporal image assistance. This approach can not only achieve a reliable modification to the original change detection results, but also produce two additional improved change detection results in the logical circulation of land surface change automatically. The proposed method consists of three parts: traditional CVA change detection, automated sample selection, and refined modification based on SVM posterior probability comparison in temporal space. It was experimented by land cover change detection from Sentinel-2 and Planet Labs images in three study areas located in Ma’anshan, Nanjing and Taizhou City. The results show that accuracies have significant improvements by the TLCVA approach, and omission and commission errors reduce obviously. The generalization, sensitivity and efficiency of the proposed approach were also analyzed in the experiments. It is concluded that different threshold decision methods of preliminary CVA in the proposed approach can work effectively and efficiently, and a small size of training samples selected from the automated sample decision method is enough to achieve improved change detection performance.

Good knowledge of inland water dynamics is of great significance for water management, preserving ecological balance and supporting industrial and agricultural development. However, the existing water cover products and water extraction methods cannot meet the present needs of monitoring water distribution and dynamic changes accurately and timely, particularly in the areas frequently disturbed by human activities, such as the Taihu Lake region. This article proposed an expert knowledge system to detect annual stable water and separate aquaculture water from natural water, and a frequency-based approach is used to generate stable water map within a year. All available Landsat Level-2 images were used to generate annual 30-m resolution stable water products from 1984 to 2018, and analyze the historical spatial-temporal changes of the water body in the Taihu Lake region. Furthermore, we related each important graph change with a reality event at that time. The results suggest that human activities have an obviously stronger influence on surface water than climate fluctuations in the Taihu Lake region, and confirm the effectiveness of ecological protection policy in maintaining the stability of the total amount of natural water in the past few decades. The spatial-temporal disturbance of aquaculture also provided another perspective and a reliable evidence of previous studies on the influence of human activities on the eutrophication process of Taihu Lake.

VHR remote sensing image change detection with pixel-based method often results in some problems that have negative effects on accuracy, such as the salt-and -pepper noise. In order to achieve a better result under this circumstance, an unsupervised sequential strategy combining Morphological Profiles and automated training sample extraction is introduced. Change detection with two real multi-temporal VHR datasets were carried out to test the effectiveness of the proposed approach. The experimental results showed that this approach outperformed the traditional unsupervised change detection methods in terms of accuracy and visual effect.

Hyperspectrum logging of core is one of the effective techniques for excavating geological data deeply and making a breakthrough in geological prospecting. Using hyperspectral remote sensing technology and deep rock sampling based on drilling technology has the advantages of mineral recognition. The first core scanner CMS350A in China has been developed successfully through the special project of national great scientific instruments and equipment, namely "development and marketing of core spectral scanner". In consideration of the data collecting mechanism and characteristics of acquired core images and spectra by the scanner, the authors focused on developing the data preprocessing methods for the core image and spectrum data. A radiation correction method based on standard plate was developed for core scanning image, an interference spectrum detection and modification technology was proposed, and a model for automatic core image extraction and mosaicking was created to accurately process the data in time. These methods constitute the basis for core spectrum analysis, physic - chemical parameters inversion, and mineral analysis in future.

Feature representation is a classic problem in the machine learning community due to the fact that different representations can entangle and hide more or less the different explanatory factors of variation behind the raw data. Especially for scene classification, its performance generally depends on the discriminative power of feature representation. Recently, unsupervised feature learning attracts tremendous attention because of its ability to learn feature representation automatically. However, reliable performance of feature representations by unsupervised learning always requires a large number of features and complex framework of mid-level feature representation. To alleviate such drawbacks, this paper presents a new framework of mid-level feature representation, which does not need learn many convolutional features during the unsupervised feature learning process, and has few parameter settings. In detail, the unsupervised feature learning method, sparse autoencoder, is employed to learn relatively small number of convolutional features from input dataset, and then extended features are extracted from the learned features by a multiple normalized difference features extraction method to compose a derivative feature set. At mid-level feature representation stage, in order to avoid poor performance of standard pooling technology in solving problems brought by rotation and translation of scene images, global feature descriptors (histogram moments, mean, variance, standard deviation) are utilized to build mid-level feature representations of images. For validation and comparison purposes, the proposed approach is evaluated via experiments with two challenging high-resolution remote sensing datasets. The results demonstrate that the approach is effective, and shows strong performance for remotely sensed scene classification.

In this letter, we propose a kernel fused representation-based classifier (KFRC) for hyperspectral images (HSIs), which combines sparse representation (SR) and collaborative representation (CR) into a unified kernel representation-based classification framework. First, we present two individual kernel methods, i.e., kernel SR (KSR) and kernel CR (KCR), which kernelize the representation methods by projecting the samples into a high-dimensional kernel space to improve the samples separability between different classes. Once obtaining the two kernel representation coefficients, KFRC attempts to achieve a balance between KSR and KCR via an adjusting parameter $\theta $ in the kernel residual domain. Subsequently, the class label of each test sample is determined by the minimum residual for each class. Experimental results on two HSIs demonstrate the proposed kernel fused method performs better than the other state-of-the-art representation-based classifiers.