The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences
Download
Publications Copernicus
Download
Citation
Articles | Volume XLII-3/W12-2020
Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci., XLII-3/W12-2020, 55–60, 2020
https://doi.org/10.5194/isprs-archives-XLII-3-W12-2020-55-2020
Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci., XLII-3/W12-2020, 55–60, 2020
https://doi.org/10.5194/isprs-archives-XLII-3-W12-2020-55-2020

  04 Nov 2020

04 Nov 2020

ASSESSMENT OF PCA AND MNF INFLUENCE IN THE VHR SATELLITE IMAGE CLASSIFICATIONS

P. C. Molina1, M. P. Castro1, and C. S. Anjos2 P. C. Molina et al.
  • 1Surveying and Cartography Engineering student, Federal Institute of Education, Science and Technology of the South of Minas Gerais, Inconfidentes, Minas Gerais, Brazil
  • 2Dept. of Surveying and Cartography, Federal Institute of Education, Science and Technology of the South of Minas Gerais, Inconfidentes, Minas Gerais, Brazil

Keywords: Classification Accuracy, Spectral Transformations, Random Forest, Deep Learning, Remote Sensing, Satellite Imaging

Abstract. Orbital images have been increasingly refined spatially as spectrally as that is the case with those provided by satellite Earth observation WorldView-3 used in this paper. However, the images are very susceptible to noise interference, so it is difficult to identify and characterize objects. Therefore, it is essential to use techniques to minimize them. Thus, through increasingly innovative processing, it is possible to carry out detailed characterization mainly of urban areas. This work aims to perform the classification of images Worldview-3 using the advanced methods of classification Random Forest and Deep Learning for the region of Botafogo in the municipality of Rio de Janeiro, Brazil. Such classifications were performed for four different data sets, including the spectral bands and transformations (MNF and PCA) resulting from the original images. The results demonstrate that the use of transformations resulting from the original images as input data for the extraction of attributes in conjunction with the spectral bands improves the accuracy of the classifications generated by the Random Forest and Deep Learning method.