Injecting spectral indices to transferable convolutional neural network under imbalanced and noisy labels for Landsat image classification
Stable and continuous remote sensing land-cover mapping is important for agriculture, ecosystems, and land management. Convolutional neural networks (CNNs) are promising methods for achieving this goal. However, the large number of high-quality training samples required to train a CNN is difficult t...
Main Authors: | , , , , |
---|---|
Format: | Article |
Language: | English |
Published: |
Taylor & Francis Group
2022-12-01
|
Series: | International Journal of Digital Earth |
Subjects: | |
Online Access: | http://dx.doi.org/10.1080/17538947.2022.2036833 |
_version_ | 1797678484310458368 |
---|---|
author | Xuemei Zhao Jun Wu Haijian Wang Xingyu Gao Longlong Zhao |
author_facet | Xuemei Zhao Jun Wu Haijian Wang Xingyu Gao Longlong Zhao |
author_sort | Xuemei Zhao |
collection | DOAJ |
description | Stable and continuous remote sensing land-cover mapping is important for agriculture, ecosystems, and land management. Convolutional neural networks (CNNs) are promising methods for achieving this goal. However, the large number of high-quality training samples required to train a CNN is difficult to acquire. In practice, imbalanced and noisy labels originating from existing land-cover maps can be used as alternatives. Experiments have shown that the inconsistency in the training samples has a significant impact on the performance of the CNN. To overcome this drawback, a method is proposed to inject highly consistent information into the network, to learn general and transferable features to alleviate the impact of imperfect training samples. Spectral indices are important features that can provide consistent information. These indices can be fused with CNN feature maps which utilize information entropy to choose the most appropriate CNN layer, to compensate for the inconsistency caused by the imbalanced, noisy labels. The proposed transferable CNN, tested with imbalanced and noisy labels for inter-regional Landsat time-series, not only is superior in terms of accuracy for land-cover mapping but also demonstrates excellent transferability between regions in both time series and cross-regional Landsat image classification. |
first_indexed | 2024-03-11T23:00:25Z |
format | Article |
id | doaj.art-5187ac49525546319335f1de882dda79 |
institution | Directory Open Access Journal |
issn | 1753-8947 1753-8955 |
language | English |
last_indexed | 2024-03-11T23:00:25Z |
publishDate | 2022-12-01 |
publisher | Taylor & Francis Group |
record_format | Article |
series | International Journal of Digital Earth |
spelling | doaj.art-5187ac49525546319335f1de882dda792023-09-21T14:57:10ZengTaylor & Francis GroupInternational Journal of Digital Earth1753-89471753-89552022-12-0115143746210.1080/17538947.2022.20368332036833Injecting spectral indices to transferable convolutional neural network under imbalanced and noisy labels for Landsat image classificationXuemei Zhao0Jun Wu1Haijian Wang2Xingyu Gao3Longlong Zhao4Guilin University of Electronic TechnologyGuilin University of Electronic TechnologyGuilin University of Electronic TechnologyGuilin University of Electronic TechnologyChinese Academy of SciencesStable and continuous remote sensing land-cover mapping is important for agriculture, ecosystems, and land management. Convolutional neural networks (CNNs) are promising methods for achieving this goal. However, the large number of high-quality training samples required to train a CNN is difficult to acquire. In practice, imbalanced and noisy labels originating from existing land-cover maps can be used as alternatives. Experiments have shown that the inconsistency in the training samples has a significant impact on the performance of the CNN. To overcome this drawback, a method is proposed to inject highly consistent information into the network, to learn general and transferable features to alleviate the impact of imperfect training samples. Spectral indices are important features that can provide consistent information. These indices can be fused with CNN feature maps which utilize information entropy to choose the most appropriate CNN layer, to compensate for the inconsistency caused by the imbalanced, noisy labels. The proposed transferable CNN, tested with imbalanced and noisy labels for inter-regional Landsat time-series, not only is superior in terms of accuracy for land-cover mapping but also demonstrates excellent transferability between regions in both time series and cross-regional Landsat image classification.http://dx.doi.org/10.1080/17538947.2022.2036833landsat image classificationimbalanced and noisy labelconvolutional neural network (cnn)transferabilityfeature fusioninformation entropy |
spellingShingle | Xuemei Zhao Jun Wu Haijian Wang Xingyu Gao Longlong Zhao Injecting spectral indices to transferable convolutional neural network under imbalanced and noisy labels for Landsat image classification International Journal of Digital Earth landsat image classification imbalanced and noisy label convolutional neural network (cnn) transferability feature fusion information entropy |
title | Injecting spectral indices to transferable convolutional neural network under imbalanced and noisy labels for Landsat image classification |
title_full | Injecting spectral indices to transferable convolutional neural network under imbalanced and noisy labels for Landsat image classification |
title_fullStr | Injecting spectral indices to transferable convolutional neural network under imbalanced and noisy labels for Landsat image classification |
title_full_unstemmed | Injecting spectral indices to transferable convolutional neural network under imbalanced and noisy labels for Landsat image classification |
title_short | Injecting spectral indices to transferable convolutional neural network under imbalanced and noisy labels for Landsat image classification |
title_sort | injecting spectral indices to transferable convolutional neural network under imbalanced and noisy labels for landsat image classification |
topic | landsat image classification imbalanced and noisy label convolutional neural network (cnn) transferability feature fusion information entropy |
url | http://dx.doi.org/10.1080/17538947.2022.2036833 |
work_keys_str_mv | AT xuemeizhao injectingspectralindicestotransferableconvolutionalneuralnetworkunderimbalancedandnoisylabelsforlandsatimageclassification AT junwu injectingspectralindicestotransferableconvolutionalneuralnetworkunderimbalancedandnoisylabelsforlandsatimageclassification AT haijianwang injectingspectralindicestotransferableconvolutionalneuralnetworkunderimbalancedandnoisylabelsforlandsatimageclassification AT xingyugao injectingspectralindicestotransferableconvolutionalneuralnetworkunderimbalancedandnoisylabelsforlandsatimageclassification AT longlongzhao injectingspectralindicestotransferableconvolutionalneuralnetworkunderimbalancedandnoisylabelsforlandsatimageclassification |