Classifying Land Cover from Satellite Images Using Time Series Analytics

The Earth’s surface is continuously observed by satellites, leading to large multi-spectral image data sets of increasing spatial resolution and temporal density. One important application of satellite data is the mapping of land cover and land use changes such as urbanization, deforestation, and desertification. This information should be obtained automatically, with high accuracy, and at the pixel level, which implies the need to classify millions of pixels even when only small regions are studied. Balancing runtime and accuracy for this task becomes even more challenging with the recent availability of multiple time points per pixel, created by periodically performed satellite scans. In this paper we describe a novel approach to classify land cover from series of multi-spectral satellite images based on multivariate time series analytics. The main advantage of our method is that it inherently models the periodic changes (seasons, agriculture etc.) underlying many types of land covers and that it is comparably robust to noise. Compared to a classical feature-based classifier, our new method shows a slightly superior overall accuracy, with an increase of up to 20% in accuracy for rare land cover classes, though at the cost of notably increased runtime. The highest accuracy is achieved by combining both approaches.

[1]  S. Linden,et al.  Mapping pasture management in the Brazilian Amazon from dense Landsat time series , 2018 .

[2]  Yuan Li,et al.  Rotation-invariant similarity in time series using bag-of-patterns representation , 2012, Journal of Intelligent Information Systems.

[3]  Dirk Pflugmacher,et al.  Time Series Analyses in a New Era of Optical Satellite Data , 2015 .

[4]  G. Hay,et al.  An automated object-based approach for the multiscale image segmentation of forest scenes , 2005 .

[5]  Cornelius Senf,et al.  Mapping land cover in complex Mediterranean landscapes using Landsat: Improved classification accuracies from integrating multi-seasonal and synthetic imagery , 2015 .

[6]  James Large,et al.  The Great Time Series Classification Bake Off: An Experimental Evaluation of Recently Proposed Algorithms. Extended Version , 2016, ArXiv.

[7]  Andy Liaw,et al.  Classification and Regression by randomForest , 2007 .

[8]  Xingfa Gu,et al.  Land cover classification using Landsat 8 Operational Land Imager data in Beijing, China , 2014 .

[9]  Ulf Leser,et al.  Multivariate Time Series Classification with WEASEL+MUSE , 2017, ArXiv.

[10]  Limin Yang,et al.  Development of a 2001 National land-cover database for the United States , 2004 .

[11]  C. Woodcock,et al.  Continuous change detection and classification of land cover using all available Landsat data , 2014 .

[12]  Lorenzo Bruzzone,et al.  A Review of Modern Approaches to Classification of Remote Sensing Data , 2014 .

[13]  J. R. Quinlan.  Induction of decision trees , 2004, Machine Learning.

[14]  W. Keeton,et al.  Forest cover change and illegal logging in the Ukrainian Carpathians in the transition period from 1988 to 2007 , 2009 .

[15]  Eamonn J. Keogh,et al.  Searching and Mining Trillions of Time Series Subsequences under Dynamic Time Warping , 2012, KDD.

[16]  A. S. Belward,et al.  Who launched what, when and why; trends in global land-cover observation capacity from civilian earth observation satellites , 2015 .

[17]  Dirk Pflugmacher,et al.  Mapping Annual Land Use and Land Cover Changes Using MODIS Time Series , 2014, IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing.

[18]  Floriana Esposito,et al.  End-to-end Learning of Deep Spatio-temporal Representations for Satellite Image Time Series Classification , 2017, DC@PKDD/ECML.

[19]  Lars Schmidt-Thieme,et al.  Ultra-Fast Shapelets for Time Series Classification , 2015, ArXiv.

[20]  Scott L. Powell,et al.  Quantification of impervious surface in the Snohomish Water Resources Inventory Area of Western Washington from 1972–2006 , 2007 .

[21]  Dirk Pflugmacher,et al.  Cross-border forest disturbance and the role of natural rubber in mainland Southeast Asia using annual Landsat time series , 2015 .

[22]  Patrick Schäfer.  The BOSS is concerned with time series classification in the presence of noise , 2014, Data Mining and Knowledge Discovery.

[23]  Ulf Leser,et al.  Benchmarking Univariate Time Series Classifiers , 2017, BTW.

[24]  Ulf Leser,et al.  Fast and Accurate Time Series Classification with WEASEL , 2017, CIKM.

[25]  Steven W. Running,et al.  Ecosystem Disturbance, Carbon, and Climate , 2008, Science.

[26]  Thomas R. Loveland,et al.  A review of large area monitoring of land cover change using Landsat data , 2012 .

[27]  Jianwen Ma,et al.  Land cover classification from MODIS EVI times-series data using SOM neural network , 2005 .

[28]  George C. Runger,et al.  Time series representation and similarity based on local autopatterns , 2016, Data Mining and Knowledge Discovery.