Performance Analysis of Deep Convolutional Autoencoders with Different Patch Sizes for Change Detection from Burnt Areas
Fire is one of the primary sources of damages to natural environments globally. Estimates show that approximately 4 million km<sup>2</sup> of land burns yearly. Studies have shown that such estimates often underestimate the real extent of burnt land, which highlights the need to find bet...
| Main Authors: | , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
MDPI AG
2020-08-01
|
| Series: | Remote Sensing |
| Subjects: | |
| Online Access: | https://www.mdpi.com/2072-4292/12/16/2576 |
| _version_ | 1827710452027621376 |
|---|---|
| author | Pablo Pozzobon de Bem Osmar Abílio de Carvalho Júnior Osmar Luiz Ferreira de Carvalho Roberto Arnaldo Trancoso Gomes Renato Fontes Guimarães |
| author_facet | Pablo Pozzobon de Bem Osmar Abílio de Carvalho Júnior Osmar Luiz Ferreira de Carvalho Roberto Arnaldo Trancoso Gomes Renato Fontes Guimarães |
| author_sort | Pablo Pozzobon de Bem |
| collection | DOAJ |
| description | Fire is one of the primary sources of damages to natural environments globally. Estimates show that approximately 4 million km<sup>2</sup> of land burns yearly. Studies have shown that such estimates often underestimate the real extent of burnt land, which highlights the need to find better, state-of-the-art methods to detect and classify these areas. This study aimed to analyze the use of deep convolutional Autoencoders in the classification of burnt areas, considering different sample patch sizes. A simple Autoencoder and the U-Net and ResUnet architectures were evaluated. We collected Landsat 8 OLI+ data from three scenes in four consecutive dates to detect the changes specifically in the form of burnt land. The data were sampled according to four different sampling strategies to evaluate possible performance changes related to sampling window sizes. The training stage used two scenes, while the validation stage used the remaining scene. The ground truth change mask was created using the Normalized Burn Ratio (<i>NBR</i>) spectral index through a thresholding approach. The classifications were evaluated according to the <i>F</i>1 index, <i>Kappa</i> index, and mean Intersection over Union (<i>mIoU</i>) value. Results have shown that the U-Net and ResUnet architectures offered the best classifications with average <i>F</i>1, <i>Kappa</i>, and <i>mIoU</i> values of approximately 0.96, representing excellent classification results. We have also verified that a sampling window size of 256 by 256 pixels offered the best results. |
| first_indexed | 2024-03-10T17:39:17Z |
| format | Article |
| id | doaj.art-0477a7cdbd45467c85e0a8c2c67a9cc3 |
| institution | Directory Open Access Journal |
| issn | 2072-4292 |
| language | English |
| last_indexed | 2024-03-10T17:39:17Z |
| publishDate | 2020-08-01 |
| publisher | MDPI AG |
| record_format | Article |
| series | Remote Sensing |
| spelling | doaj.art-0477a7cdbd45467c85e0a8c2c67a9cc32023-11-20T09:44:26ZengMDPI AGRemote Sensing2072-42922020-08-011216257610.3390/rs12162576Performance Analysis of Deep Convolutional Autoencoders with Different Patch Sizes for Change Detection from Burnt AreasPablo Pozzobon de Bem0Osmar Abílio de Carvalho Júnior1Osmar Luiz Ferreira de Carvalho2Roberto Arnaldo Trancoso Gomes3Renato Fontes Guimarães4Departamento de Geografia, Campus Universitário Darcy Ribeiro, Asa Norte, Universidade de Brasília, DF, Brasília 70910-900, BrazilDepartamento de Geografia, Campus Universitário Darcy Ribeiro, Asa Norte, Universidade de Brasília, DF, Brasília 70910-900, BrazilDepartamento de Engenharia Elétrica, Campus Universitário Darcy Ribeiro, Asa Norte, Universidade de Brasília, DF, Brasília 70910-900, BrazilDepartamento de Geografia, Campus Universitário Darcy Ribeiro, Asa Norte, Universidade de Brasília, DF, Brasília 70910-900, BrazilDepartamento de Geografia, Campus Universitário Darcy Ribeiro, Asa Norte, Universidade de Brasília, DF, Brasília 70910-900, BrazilFire is one of the primary sources of damages to natural environments globally. Estimates show that approximately 4 million km<sup>2</sup> of land burns yearly. Studies have shown that such estimates often underestimate the real extent of burnt land, which highlights the need to find better, state-of-the-art methods to detect and classify these areas. This study aimed to analyze the use of deep convolutional Autoencoders in the classification of burnt areas, considering different sample patch sizes. A simple Autoencoder and the U-Net and ResUnet architectures were evaluated. We collected Landsat 8 OLI+ data from three scenes in four consecutive dates to detect the changes specifically in the form of burnt land. The data were sampled according to four different sampling strategies to evaluate possible performance changes related to sampling window sizes. The training stage used two scenes, while the validation stage used the remaining scene. The ground truth change mask was created using the Normalized Burn Ratio (<i>NBR</i>) spectral index through a thresholding approach. The classifications were evaluated according to the <i>F</i>1 index, <i>Kappa</i> index, and mean Intersection over Union (<i>mIoU</i>) value. Results have shown that the U-Net and ResUnet architectures offered the best classifications with average <i>F</i>1, <i>Kappa</i>, and <i>mIoU</i> values of approximately 0.96, representing excellent classification results. We have also verified that a sampling window size of 256 by 256 pixels offered the best results.https://www.mdpi.com/2072-4292/12/16/2576deep learningCNNclassificationfiremultitemporal image |
| spellingShingle | Pablo Pozzobon de Bem Osmar Abílio de Carvalho Júnior Osmar Luiz Ferreira de Carvalho Roberto Arnaldo Trancoso Gomes Renato Fontes Guimarães Performance Analysis of Deep Convolutional Autoencoders with Different Patch Sizes for Change Detection from Burnt Areas Remote Sensing deep learning CNN classification fire multitemporal image |
| title | Performance Analysis of Deep Convolutional Autoencoders with Different Patch Sizes for Change Detection from Burnt Areas |
| title_full | Performance Analysis of Deep Convolutional Autoencoders with Different Patch Sizes for Change Detection from Burnt Areas |
| title_fullStr | Performance Analysis of Deep Convolutional Autoencoders with Different Patch Sizes for Change Detection from Burnt Areas |
| title_full_unstemmed | Performance Analysis of Deep Convolutional Autoencoders with Different Patch Sizes for Change Detection from Burnt Areas |
| title_short | Performance Analysis of Deep Convolutional Autoencoders with Different Patch Sizes for Change Detection from Burnt Areas |
| title_sort | performance analysis of deep convolutional autoencoders with different patch sizes for change detection from burnt areas |
| topic | deep learning CNN classification fire multitemporal image |
| url | https://www.mdpi.com/2072-4292/12/16/2576 |
| work_keys_str_mv | AT pablopozzobondebem performanceanalysisofdeepconvolutionalautoencoderswithdifferentpatchsizesforchangedetectionfromburntareas AT osmarabiliodecarvalhojunior performanceanalysisofdeepconvolutionalautoencoderswithdifferentpatchsizesforchangedetectionfromburntareas AT osmarluizferreiradecarvalho performanceanalysisofdeepconvolutionalautoencoderswithdifferentpatchsizesforchangedetectionfromburntareas AT robertoarnaldotrancosogomes performanceanalysisofdeepconvolutionalautoencoderswithdifferentpatchsizesforchangedetectionfromburntareas AT renatofontesguimaraes performanceanalysisofdeepconvolutionalautoencoderswithdifferentpatchsizesforchangedetectionfromburntareas |