Analysing Satellite Image Time Series by Means of Pattern Mining

2011

The interpretation of remotely sensed images in a spatiotemporal context is becoming a valuable research topic. However, the constant growth of data volume in remote sensing imaging makes reaching conclusions based on collected data a challenging task. Recently, data mining appears to be a promising research field leading to several interesting discoveries in various areas such as marketing, surveillance, fraud detection and scientific discovery. By integrating data mining and image interpretation techniques, accurate and relevant information (i.e. functional relation between observed parcels and a set of informational contents) can be automatically elicited.

IGARSS 2008 - 2008 IEEE International Geoscience and Remote Sensing Symposium, 2008

Nowadays, there is a growing need for processing huge volumes of observation data due to the increase in size, in resolution, in spectral channel number and in acquisition frequency of remote sensing images. When data is gathered over time for a same geographical zone, this data is said to be a Satellite Image Time Series (SITS). The informational content of SITS is rich because the observed scene is described both in time and in space. In order to exhibit potential interesting spatio-temporal patterns, we propose to extract pixel-based evolutions from SITS data by using two different symbolic techniques. The first one is based on data mining techniques that aim at extracting frequent sequential patterns (e.g., ). The second one relies on the use of tries (e.g., [2], [3]) for classifying pixels according to their evolution in time. Encouraging experiments on a SPOT SITS are detailed.

2010

A wealth of remotely sensed image time series covering large areas is now available to the earth science community. Change detection methods are often not capable of detecting land cover changes within time series that are heavily influenced by seasonal climatic variations. Detecting change within the trend and seasonal components of time series enables the classification of different types of changes.

Journal of the Indian Society of Remote Sensing

Change detection through the analysis of images is a fundamental step in the remote sensing analysis framework. It is an emerging area of research with many methods and algorithms in place. Most of the work till date is based on applied change detection methods to a pair of images in the case of satellite imagery. This paper proposes a novel change point detection methodology for a time-ordered set of images which shall enhance the efficiency within the overall image processing framework. It focuses on change detection applied to the set of time-ordered images to identify the exact pair of bi-temporal images about the change point, thereby being of great value to the image analyst in the overall image processing workflow. The paper proposes a metric to detect changes in time-ordered image series in the form of rank ordered threshold values extracted post-application of the segmentation algorithms to the bi-temporal image pair differences derived from time-ordered images. The rank of the threshold value specific to the image pair indicates the relevance and quantum of changes in the image pair among the entire time-ordered image series. A higher rank of threshold conforms to the pair of images with higher image changes. A total of four segmentation algorithms including the wellknown Otsu, minimum cross-entropy method by Li et al. methods have been applied to a set of ten time-ordered satellite image sequences as part of the study. Li's cross-entropy method is found to provide the best results for the determination of the change point. Such a change point detection methodology is applicable not only to satellite imagery but also to a general time-ordered set of images and video frames.

Proceedings of the 16th International Conference on Enterprise Information Systems, 2014

The amount of data generated and stored in many domains has increased in the last years. In remote sensing, this scenario of bursting data is not different. As the volume of satellite images stored in databases grows, the demand for computational algorithms that can handle and analyze this volume of data and extract useful patterns has increased. In this context, the computational support for satellite images data analysis becomes essential. In this work, we present the SITSMining framework, which applies a methodology based on data mining techniques to extract patterns and information from time series obtained from satellite images. In Brazil, as the agricultural production provides great part of the national resources, the analysis of satellite images is a valuable way to help crops monitoring over seasons, which is an important task to the economy of the country. Thus, we apply the framework to analyze multitemporal satellite images, aiming to help crop monitoring and forecasting of Brazilian agriculture.

IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2011

An automated land cover change detection method is proposed that uses coarse spatial resolution hyper-temporal earth observation satellite time series data. The study compared three different unsupervised clustering approaches that operate on short term Fourier transform coefficients computed over subsequences of 8-day composite MODerate-resolution Imaging Spectroradiometer (MODIS) surface reflectance data that were extracted with a temporal sliding window. The method uses a feature extraction process that creates meaningful sequential time series that can be analyzed and processed for change detection. The method was evaluated on real and simulated land cover change examples and obtained a change detection accuracy exceeding 76% on real land cover conversion and more than 70% on simulated land cover conversion.

IEEE Geoscience and Remote Sensing Letters, 2009

In this letter, we propose a novel technique for unsupervised change detection in multitemporal satellite images using principal component analysis (PCA) and k-means clustering. The difference image is partitioned into h × h nonoverlapping blocks. S, S ≤ h 2 , orthonormal eigenvectors are extracted through PCA of h × h nonoverlapping block set to create an eigenvector space. Each pixel in the difference image is represented with an S-dimensional feature vector which is the projection of h × h difference image data onto the generated eigenvector space. The change detection is achieved by partitioning the feature vector space into two clusters using k-means clustering with k = 2 and then assigning each pixel to the one of the two clusters by using the minimum Euclidean distance between the pixel's feature vector and mean feature vector of clusters. Experimental results confirm the effectiveness of the proposed approach.

Sādhanā

Land cover change detection has been a topic of active research in the remote sensing community. Due to enormous amount of data available from satellites, it has attracted the attention of data mining researchers to search a new direction for solution. The Terra Moderate Resolution Imaging Spectrometer (MODIS) vegetation index (EVI/NDVI) data products are used for land cover change detection. These data products are associated with various challenges such as seasonality of data, spatio-temporal correlation, missing values, poor quality measurement, high resolution and high dimensional data. The land cover change detection has often been performed by comparing two or more satellite snapshot images acquired on different dates. The image comparison techniques have a number of limitations. The data mining technique addresses many challenges such as missing value and poor quality measurements present in the data set, by performing the preprocessing of data. Furthermore, the data mining approaches are capable of handling large data sets and also use some of the inherent characteristics of spatio-temporal data; hence, they can be applied to increasingly immense data set. This paper stretches in detail various data mining algorithms for land cover change detection and each algorithm's advantages and limitations. Also, an empirical study of some existing land cover change detection algorithms and results have been presented in this paper.

2006

The frequency and the resolution of the images acquired by remote sensing techniques are nowadays so high that end-users can get huge volumes of observation data for a same geographic area. In this paper, we propose to use data mining tools for detecting evolutions. More precisely, we explain how to make use of sequential patterns to automatically extract evolutions that are contained in a satellite images series which is considered as a base of sequences. Experiments on optical data from METEOSAT satellite and on Synthetic Aperture Radar (SAR) images from European Remote Sensing (ERS) satellites are presented.

2009

Change detection analyze means that according to observations made in different times, the process of defining the change detection occurring in nature or in the state of any objects or the ability of defining the quantity of temporal effects by using multitemporal data sets. There are lots of change detection techniques met in literature. It is possible to group these techniques under two main topics as supervised and unsupervised change detection. In this study, the aim is to define the land cover changes occurring in specific area of Kayseri with unsupervised change detection techniques by using Landsat satellite images belonging to different years which are obtained by the technique of remote sensing. While that process is being made, image differencing method is going to be applied to the images by following the procedure of image enhancement. After that, the method of Principal Component Analysis is going to be applied to the difference image obtained. To determine the areas that have and don't have changes, the image is grouped as two parts by Fuzzy C-Means Clustering method. For achieving these processes, firstly the process of image to image registration is completed. As a result of this, the images are being referred to each other. After that, gray scale difference image obtained is partitioned into 3x3 nonoverlapping blocks. With the method of principal component analysis, eigenvector space is gained and from here, principal components are reached. Finally, feature vector space consisting principal component is partitioned into two clusters using Fuzzy C-Means Clustering and after that change detection process has been done.

2011

Satellite Image Time Series (SITS) analysis is an important domain with various applications in land study. In the coming years, both high temporal and high spatial resolution SITS will be available. This article aims at providing both temporal and spatial analysis of SITS. We propose first segmenting each image of the series, and then using these segmentations in order to characterize each pixel of the data with a spatial dimension (i.e. with contextual information). Providing spatially characterized pixels, pixel-based temporal analysis can be performed. Experiments carried out with this methodology show the relevance of this approach and the significance of the resulting extracted patterns in the context of the analysis of SITS.

2007 International Workshop on the Analysis of Multi-temporal Remote Sensing Images, 2007

Very recently satellite systems for remote sensing are required to provide images with a spatial and temporal resolution suitable to be applied for disaster management. High resolution (HR) satellite imagery can provide a good insight into the magnitude of a disaster and a detailed assessment of the damage. To meet these objectives, HR imagery has to be collected immediately after the disaster and precisely in the areas that have been damaged by the event. Presently, space based remote sensing systems result unsuitable to provide useful information when disastrous events require simultaneously high temporal and spatial resolutions. Furthermore, due to the technological limits of the transmission systems, a very high resolution is usually coupled with a reduced sensor swath. This means that the observation can be carried out when the area to be imaged is known. Low-resolution satellites (e.g. geostationary satellite) could also provide, in principle, some information with the required promptness in presence of event characterized by sudden temperature increases (fires, explosions, volcanic eruption, etc). The University of Rome (Centro di Ricerca Progetto San Marco) is studying the suitability of a satellite based system able to monitor national borders and/or given regions of the Earth in a quasi-continuous way with an adequate spatial resolution. To meet this requirement, the so-called Multi-Stationary (MS) orbits have been introduced. A constellation of few (4) satellites located on this kind of orbits allows a quasi-continuous monitoring of a selected region of the Earth. This paper is devoted to assess the impact of the variability of the images spatial resolution and illumination conditions on change detection methods based on a time-series of images.

2011

The majority of phenological studies have focussed on extracting critical points, i.e. phenological metrics such as start-of-season, in the seasonal growth cycle. These metrics do not exploit the full temporal detail of time series, depend on their definition or threshold, and are influenced by disturbances. Here, we evaluated a robust phenological change detection ability of a method for detecting abrupt, gradual, and phenological changes within time series. BFAST, Breaks For Additive Seasonal and Trend method, integrates the decomposition of time series into trend, seasonal, and remainder components with methods for detecting change within trend and seasonal (i.e. phenology) component. We tested BFAST by analysing 16-day MODIS NDVI composites (MOD13C1 collection 5) between 2000-2009 covering Australia. This illustrated that the method is able to detect the timing of major phenological changes within time series while accounting for abrupt disturbances and gradual trends. It was also shown that the phenological change detection is influenced by the signal-to-noise ratio of the time series. The BFAST method is a generic change detection method which can be applied to any time series data. The methods are available in the BFAST package for R from CRAN (http://CRAN.R-project. org/package=bfast).

Currently remote sensing, based on satellite images is one of the most important source of information for multitemporal change detection. From all types of satellite images, the multispectral images present the advantage of characterizing the earth surface in different bands; each band provides different and useful information. In this work we propose a new methodology based on linear PCA to extract useful and meaningful information from signals provided by the remote sensing, and based on it, detect temporal changes Experiments based on images of the satellite CBERS-2B corresponding to the urban and peri urban region of Rio Cuarto of Córdoba state in Argentina have given satisfactory results in change detection.

Journal of Electrical and Computer Engineering, 2017

Change detection (CD) of any surface using multitemporal remote sensing images is an important research topic since up-to-date information about earth surface is of great value. Abrupt changes are occurring in different earth surfaces due to natural disasters or man-made activities which cause damage to that place. Therefore, it is necessary to observe the changes for taking necessary steps to recover the subsequent damage. This paper is concerned with this issue and analyzes statistical similarity measure to perform CD using remote sensing images of the same scene taken at two different dates. A variation of normalized mutual information (NMI) as a similarity measure has been developed here using sliding window of different sizes. In sliding window approach, pixels' local neighborhood plays a significant role in computing the similarity compared to the whole image. Thus the insignificant global characteristics containing noise and sparse samples can be avoided when evaluating the probability density function. Therefore, NMI with different window sizes is proposed here to identify changes using multitemporal data. Experiments have been carried out using two separate multitemporal remote sensing images captured one year apart and one month apart, respectively. Experimental analysis reveals that the proposed technique can detect up to 97.71% of changes which outperforms the traditional approaches.