Detection of clearing areas in the Amazon Forest through satellite monitoring using machine learning techniques
Abstract
This paper addresses the problem of clearing detection in the Amazon rainforest using remote sensing data and (un)supervised machine learning techniques. A clearing (or canopy gap) is a small area in a forest where there are no trees, or there is only low-lying forest which differs from its surroundings. Experiments carried out with 44 satellite images, divided into 3,288 segments, evaluated by experts, where 1,652 of non-clearing class and 1,636 of clearing. The segments were represented by a set of features that include first and second order statistics and color information. In the supervised learning approach, the best results were obtained with the rule-based methods, Decision Tree and Random Forest, reaching 97% in both classes. Methods that use different approaches have got worse results, which suggests the need for a better analysis of the relation between the attributes. Among the unsupervised learning techniques, the best performance was achieved by the BIRCH method, which obtained 94.48% for both classes. However it required a higher number of clusters under higher execution time. Preliminary results indicate that the proposed approach is promising, which must continue to be investigated.
References
Braga, A. L. and Figueiredo, G. C., S. F. S. V. P. (2006). Comparação entre as classificações híbrida e supervisionada no mapeamento do uso do solo usando imagens de alta resolução. Anais do Congresso Brasileiro de Cadastro Técnico Multifinalitário.
Cavalcanti L., A. M., Carvalho J., H., and Miranda dos Santos, E. (2015). A comparison on supervised machine learning classification techniques for semantic segmentation of aerial images of rain forest regions. Proceedings of the 10th International Conference on Computer Vision Theory and Applications (VISAPP-2015).
Curran, P. J. (1988). The semi-variogram in remote sensing: an introduction.
Dell’Acqua, F. and Gamba, P. (2003). Texture-based characterization of urban environments on satellite sar images. IEEE Trans. on Geoscience and Remote Sensing, 41(1):153–159.
Du, P., Samat, A., Waske, B., Liu, S., and Li, Z. (2015). Random forest and rotation forest for fully polarized sar image classification using polarimetric and spatial features. Journal of Photogrammetry and Remote Sensing (ISPRS), pages 38–53.
Fernandes, C. d. E. (2008). Classificação de imagens de sensoriamento remoto com áreas desmatadas.
He, C., Li, S., Liao, Z., and Liao, M. (2013). Texture classification of polsar data based on sparse coding of wavelet polarization textons. IEEE Transactions on Geoscience and Remote Sensing, 51(8):4576–4590.
Jin, H., Mountrakis, G., and Stehman, S. V. (2014). Assessing integration of intensity, polarimetric scattering, interferometric coherence and spatial texture metrics in PALSAR-derived land cover classification. ISPRS Journal of Photogrammetry and Remote Sensing, 98:70–84.
Kumar, M. and Miller, D. A. (2006). Non-parametric classification strategy for remotely sensed images using both spectral and textural information. International Multi- Conference Signal Processing, Pattern Recognition (IASTED), pages 81–89.
Magnabosco, D., Chambers, J., Higuchi, N., Trumbore, S., Ribeiro, G. H., dos Santos, J., Negr´on-Ju´arez, R., Reu, B., and Wirth, C. (2014). Large-scale wind disturbances promote tree diversity in a central amazon forest. PLoS ONE, 9:e103711.
MapBiomas (2015). Projeto de mapeamento anual da cobertura e uso do solo no brasil. projeto mapbiomas – coleção 3.1 da série anual de mapas de cobertura e uso de solo do brasil. disponível em: http://mapbiomas.org. acesso em 2019.
Olofsson, P., Foody, G., Herold, M., Stehman, S., Woodcock, C., and Wulder, M. (2014). Good practices for estimating area and assessing accuracy of land change. Remote Sensing of Environment, page 42–57.
Pla, F.; Carmona, P. L. S. J. M. (2013). One-class classification techniques in image recognition problems. Information Optics (WIO), pages 1–3.
Qi, Z., Yeh, A. G.-O., Li, X., and Lin, Z. (2012). A novel algorithm for land use and land cover classification using RADARSAT-2 polarimetric SAR data. Remote Sensing of Environment, 118:21–39.
Russell, S. and Norvig, P. (2003). Artificial Intelligence - A Modern Approach. Prentice Hall, 2th edition.
Tan, P.-N., Steinbach, M., and Kumar, V. (2009). Introdução ao Data Mining. Ciência Moderna, 1th edition.
Uhlmann, S. and Kiranyaz, S. (2014). Classification of dual- and single polarized SAR images by incorporating visual features. ISPRS Journal of Photogrammetry and Remote Sensing, 90:10–22.
Wang, X. e. a. (2013). A one-class classification by spatial-contextual for remotely sensed image. eoscience and Remote Sensing Symposium (IGARSS), page 437–440.
Witten, I. H. and Frank (2005). Data Mining – Practical Machine Learning Tools and Techniques. McGraw- Hill.
