SPIE Membership Get updates from SPIE Newsroom
  • Newsroom Home
  • Astronomy
  • Biomedical Optics & Medical Imaging
  • Defense & Security
  • Electronic Imaging & Signal Processing
  • Illumination & Displays
  • Lasers & Sources
  • Micro/Nano Lithography
  • Nanotechnology
  • Optical Design & Engineering
  • Optoelectronics & Communications
  • Remote Sensing
  • Sensing & Measurement
  • Solar & Alternative Energy
  • Sign up for Newsroom E-Alerts
  • Information for:
SPIE Photonics West 2018 | Call for Papers

SPIE Defense + Commercial Sensing 2018 | Call for Papers




Print PageEmail PageView PDF

Remote Sensing

Satellite-based multitemporal-change detection in urban environments

High-resolution optical satellite sensors can contribute to improved coastal and land management, hazard mitigation, emergency response, and ecosystem-service design.
23 February 2011, SPIE Newsroom. DOI: 10.1117/2.1201101.003502

Urbanization is defined as the physical growth of urban areas.1 Improving land management depends critically on the capacity of (near-)real-time monitoring of land-use/cover change, so that solutions to a range of urban/rural-interface development issues can be managed promptly and adequately. From multitemporal to rapid-change detection, both the resolution of satellite sensors and the computational capacity of the classifiers used for image processing must be well integrated. Eventually, this may lead to improvements in coastal and land management, hazard mitigation, emergency response, and ecosystem-service design. High-resolution satellite-based remote-sensing technology and the science associated with evaluation of land use and land cover (LULC) in urban regions use a wide range of images and algorithms for various applications.

Early remote-sensing image-classification studies employed statistical methods, such as the maximum-likelihood classifier,2 the K-nearest-neighbor algorithm,3 and the K-means clustering approach.4In recent years, methods based on artificial-intelligence and machine-learning techniques have become popular. Approaches based on neural computing, fuzzy logic, evolutionary algorithms, and expert systems are widely used.5–11 Yet, existing processing techniques using LULC methods are often time-consuming, laborious, and tedious to use, resulting in the unavailability of the results within the designated time window. We developed a new image-classification approach using 2.5m-resolution SPOT-5 (SPOT: Satellite pour l'observation de la terre/satellite for earth observation) image products to investigate coastal-land-reclamation impact, urban/rural interactions, and changes of ecosystem service in Dalian (China), a fast growing city. We constructed a classifier based on the partial Lanczos extreme-learning machine (PL-ELM), a novel algorithm with fast learning speed and outstanding generalization performance.

Our case study in the Dalian Development Area (DDA) based on images collected in 2003 and 2007 (see Figure 1) fully supports the monitoring needs and aids in rapid-change detection in terms of both urban expansion and coastal-land reclamation. Since different classes of LULC outputs may have similar spectral characteristics, we extracted texture features and vegetation indices and included them in the classification process to enhance the discernability of the results. We added more features to the image pixels and introduced the ‘normalized differential vegetation index,’ as well as four commonly used texture features (angular second moment, contrast, correlation, and homogeneity).12 Therefore, we extended the feature-space dimension for all data points/pixels to eight. A validation procedure based on ground-truth data and comparisons with a number of classic classifiers proves the credibility of our proposed PL-ELM approach in terms of both classification accuracy and processing speed. We divided the LULC features into six major categories, including water bodies, forests, grasslands, bare fields, buildings, and roads. We then compared the classification accuracies for all major types, enabling us to determine the overall accuracies as well as the Kappa coefficients. Our PL-ELM classification approach outperforms five other major algorithms, including the backward-propagation neural-network, maximum-likelihood, K-nearest-neighbor and naive Bayes' algorithms, as well as the support-vector machine.

Figure 1. Location of the Dalian Development Area (DDA) and the corresponding SPOT-5 (Satellite for earth observation-5) images from 2003 and 2007.

We assessed the classification accuracy of our approach against a test data set, which consisted of 500 ground-truth data points, and fully verified the performance of the PL-ELM classifier. Figure 2 exemplifies the land-reclamation and urban-expansion processes using these statistics. From 2003 to 2007, the area covered by water decreased by 12.7206km2 as a result of coastal-land reclamation. In addition, building coverage in the region increased by 7.8086 km2. This is strong evidence of rapid urban expansion. The rapid increase of the brown-colored area (which represents buildings) indicates rapid urbanization driven by population growth and migration. The exorbitant decrease of water coverage caused by coastal-land reclamation is a symbolic pattern of the DDA's urbanization, which is driven by the prevailing land-management policy. The drastic reduction of grassland during this short time period implies a loss of ecosystem service.

Figure 2. Final classification results produced from (a) the 2003 and (b) 2007 image sets using the partial Lanczos extreme-learning-machine classifier. LULC: Land use/land cover.

In summary, we have developed a novel LULC classification approach based on an innovative machine-learning algorithm. We applied the proposed approach to two SPOT-5 image sets, acquired in 2003 and 2007, to produce six-category LULC maps with a spatial resolution of 2.5m. To improve the prediction accuracy, we extracted additional features from the original remote-sensing images, which we took into account in the classification process. To ensure urban sustainability, multitemporal variations of grasslands, water coverage, and buildings caused by urban sprawl, coastal-land reclamation, and population increase and migration, respectively, collectively reflect the main direction and connection of the urbanization process, which traces the impact of economic development. Our results will allow development of a full-scale rapid-change detection system to monitor urbanization driven by both human actions and natural disasters.

Ni-Bin Chang
Department of Civil and Environmental Engineering University of Central Florida
Orlando, FL

Ni-Bin Chang is a professor.

1. http://en.wikipedia.org/wiki/Urbanization Wikipedia entry on urbanization. Accessed 4 January 2011.
2. P. Lombardo, C. J. Oliver, Maximum likelihood approach to the detection of changes between multitemporal SAR images, IEEE Proc. Radar Sonar Navig. 148, no. 4, pp. 200-210, 2001. doi:10.1049/ip-rsn:20010114
3. E. Blanzieri, F. Melgani, Nearest neighbor classification of remote sensing images with the maximal margin principle, IEEE Trans. Geosci. Rem. Sens. 46, no. 6, pp. 1804-1811, 2008. doi:10.1109/TGRS.2008.2010682
4. J. Zhang, Z. Cui, A. Liu, Y. Jia, A k-means remote sensing image classification method based on adaboost, IEEE Int'l Conf. Natur. Comput. (ICNC'08) 4, pp. 27-32, 2008. doi:10.1109/ICNC.2008.903
5. G. Giacinto, F. Roli, Design of effective neural network ensembles for image classification purposes, Image Vis. Comput. 19, no. 9–10, pp. 699-707, 2001. doi:10.1016/S0262-8856(01)00045-2
6. T. Kavzoglu, Increasing the accuracy of neural network classification using refined training data, Environ. Model. Softw. 24, no. 7, pp. 850-858, 2009. doi:10.1016/j.envsoft.2008.11.012
7. M. J. Canty, Boosting a fast neural network for supervised land cover classification, Comput. Geosci. 35, no. 6, pp. 1280-1295, 2009. doi:10.1016/j.cageo.2008.07.004
8. W. L. Stefanov, M. S. Ramsey, P. R. Christensen, Monitoring urban land cover change: an expert system approach to land cover classification of semiarid to arid urban centers, Rem. Sens. Environ. 77, no. 2, pp. 173-185, 2001. doi:10.1016/S0034-4257(01)00204-8
9. G. M. Foody, A. Mathur, Toward intelligent training of supervised image classifications: directing training data acquisition for SVM classification, Rem. Sens. Environ. 93, no. 1–2, pp. 107-117, 2004. doi:10.1016/j.rse.2004.06.017
10. H. W. Chen, N. B. Chang, R. F. Yu, Y. W. Huang, Urban land use and land cover classification using the neural-fuzzy inference approach with Formosat-2 data, J. Appl. Rem. Sens. 3, pp. 033558-18, 2009. doi:10.1117/1.3265995
11. B. C. K. Tso, P. M. Mather, Classification of multisource remote sensing imagery using a genetic algorithm and Markov random fields, IEEE Trans. Geosci. Rem. Sens. 37, no. 3, pp. 1255-1260, 1999. doi:10.1109/36.763284
12. J. Sklansky, Image segmentation and feature extraction, IEEE Trans. Syst. Man Cybern. 8, no. 4, pp. 237-247, 1978. doi:10.1109/TSMC.1978.4309944