Combining 3D Radiative Transfer Model and Convolutional Neural Network to Accurately Estimate Forest Canopy Cover From Very High-Resolution Satellite Images

Forest canopy cover (FCC) plays an important role in many ecological, hydrological and forestry applications. For large-scale applications, FCC is usually estimated from remotely sensed data by inverting radiative transfer models (RTMs) or using data-driven regressions. In this article, we proposed...

Descripción completa

Guardado en:
Detalles Bibliográficos
Autores principales: Decai Jin, Jianbo Qi, Huaguo Huang, Linyuan Li
Formato: article
Lenguaje:EN
Publicado: IEEE 2021
Materias:
Acceso en línea:https://doaj.org/article/85399ec30deb4e0aa5a97f8662d06943
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
id oai:doaj.org-article:85399ec30deb4e0aa5a97f8662d06943
record_format dspace
spelling oai:doaj.org-article:85399ec30deb4e0aa5a97f8662d069432021-11-18T00:00:16ZCombining 3D Radiative Transfer Model and Convolutional Neural Network to Accurately Estimate Forest Canopy Cover From Very High-Resolution Satellite Images2151-153510.1109/JSTARS.2021.3122509https://doaj.org/article/85399ec30deb4e0aa5a97f8662d069432021-01-01T00:00:00Zhttps://ieeexplore.ieee.org/document/9585661/https://doaj.org/toc/2151-1535Forest canopy cover (FCC) plays an important role in many ecological, hydrological and forestry applications. For large-scale applications, FCC is usually estimated from remotely sensed data by inverting radiative transfer models (RTMs) or using data-driven regressions. In this article, we proposed a hybrid model, which combines a 3-D RTM and transfer learning-based convolutional neural network (T-CNN), to estimate FCC from very high-resolution satellite images (e.g., Chinese GaoFen-2, 1 m resolution with 4 bands). Unlike common hybrid models that are purely trained with simulation data, T-CNN combines simulation data-based pre-training and actual data-based transfer learning, which is a widely used technique in artificial intelligence for fine-tuning models. The performance of T-CNN was compared with a random forest (RF) model and two general CNN models, including CNN trained with actual dataset only (data-CNN) and CNN trained with RTM simulation data only (RTM-CNN). Results on the independent validation dataset (not used in training stage) showed that T-CNN had higher accuracy (RMSE &#x003D; 0.121, <italic>R</italic><sup>2</sup> &#x003D; 0.83), compared with RF (RMSE &#x003D; 0.26, <italic>R</italic><sup>2</sup> &#x003D; 0.61), Data-CNN (RMSE &#x003D; 0.142, <italic>R</italic><sup>2</sup> &#x003D; 0.81), and RTM-CNN (RMSE &#x003D; 0.144, <italic>R</italic><sup>2</sup> &#x003D; 0.73), which indicates that T-CNN has a strong transferability. Tests on different training sizes showed that T-CNN (0.084 &lt; RMSE &lt; 0.108) provided constantly better performances than RF (0.116 &lt; RMSE &lt; 0.122) and data-CNN (0.103 &lt; RMSE &lt; 0.128), which demonstrates the potential of T-CNN as an alternative to RTM-based inversion and data-driven regressions to estimate FCC, especially when training data is imbalanced and inadequate.Decai JinJianbo QiHuaguo HuangLinyuan LiIEEEarticleThree-dimensional (3-D) radiative transfer modelconvolutional neural network (CNN)forest canopy cover (FCC)transfer learningOcean engineeringTC1501-1800Geophysics. Cosmic physicsQC801-809ENIEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, Vol 14, Pp 10953-10963 (2021)
institution DOAJ
collection DOAJ
language EN
topic Three-dimensional (3-D) radiative transfer model
convolutional neural network (CNN)
forest canopy cover (FCC)
transfer learning
Ocean engineering
TC1501-1800
Geophysics. Cosmic physics
QC801-809
spellingShingle Three-dimensional (3-D) radiative transfer model
convolutional neural network (CNN)
forest canopy cover (FCC)
transfer learning
Ocean engineering
TC1501-1800
Geophysics. Cosmic physics
QC801-809
Decai Jin
Jianbo Qi
Huaguo Huang
Linyuan Li
Combining 3D Radiative Transfer Model and Convolutional Neural Network to Accurately Estimate Forest Canopy Cover From Very High-Resolution Satellite Images
description Forest canopy cover (FCC) plays an important role in many ecological, hydrological and forestry applications. For large-scale applications, FCC is usually estimated from remotely sensed data by inverting radiative transfer models (RTMs) or using data-driven regressions. In this article, we proposed a hybrid model, which combines a 3-D RTM and transfer learning-based convolutional neural network (T-CNN), to estimate FCC from very high-resolution satellite images (e.g., Chinese GaoFen-2, 1 m resolution with 4 bands). Unlike common hybrid models that are purely trained with simulation data, T-CNN combines simulation data-based pre-training and actual data-based transfer learning, which is a widely used technique in artificial intelligence for fine-tuning models. The performance of T-CNN was compared with a random forest (RF) model and two general CNN models, including CNN trained with actual dataset only (data-CNN) and CNN trained with RTM simulation data only (RTM-CNN). Results on the independent validation dataset (not used in training stage) showed that T-CNN had higher accuracy (RMSE &#x003D; 0.121, <italic>R</italic><sup>2</sup> &#x003D; 0.83), compared with RF (RMSE &#x003D; 0.26, <italic>R</italic><sup>2</sup> &#x003D; 0.61), Data-CNN (RMSE &#x003D; 0.142, <italic>R</italic><sup>2</sup> &#x003D; 0.81), and RTM-CNN (RMSE &#x003D; 0.144, <italic>R</italic><sup>2</sup> &#x003D; 0.73), which indicates that T-CNN has a strong transferability. Tests on different training sizes showed that T-CNN (0.084 &lt; RMSE &lt; 0.108) provided constantly better performances than RF (0.116 &lt; RMSE &lt; 0.122) and data-CNN (0.103 &lt; RMSE &lt; 0.128), which demonstrates the potential of T-CNN as an alternative to RTM-based inversion and data-driven regressions to estimate FCC, especially when training data is imbalanced and inadequate.
format article
author Decai Jin
Jianbo Qi
Huaguo Huang
Linyuan Li
author_facet Decai Jin
Jianbo Qi
Huaguo Huang
Linyuan Li
author_sort Decai Jin
title Combining 3D Radiative Transfer Model and Convolutional Neural Network to Accurately Estimate Forest Canopy Cover From Very High-Resolution Satellite Images
title_short Combining 3D Radiative Transfer Model and Convolutional Neural Network to Accurately Estimate Forest Canopy Cover From Very High-Resolution Satellite Images
title_full Combining 3D Radiative Transfer Model and Convolutional Neural Network to Accurately Estimate Forest Canopy Cover From Very High-Resolution Satellite Images
title_fullStr Combining 3D Radiative Transfer Model and Convolutional Neural Network to Accurately Estimate Forest Canopy Cover From Very High-Resolution Satellite Images
title_full_unstemmed Combining 3D Radiative Transfer Model and Convolutional Neural Network to Accurately Estimate Forest Canopy Cover From Very High-Resolution Satellite Images
title_sort combining 3d radiative transfer model and convolutional neural network to accurately estimate forest canopy cover from very high-resolution satellite images
publisher IEEE
publishDate 2021
url https://doaj.org/article/85399ec30deb4e0aa5a97f8662d06943
work_keys_str_mv AT decaijin combining3dradiativetransfermodelandconvolutionalneuralnetworktoaccuratelyestimateforestcanopycoverfromveryhighresolutionsatelliteimages
AT jianboqi combining3dradiativetransfermodelandconvolutionalneuralnetworktoaccuratelyestimateforestcanopycoverfromveryhighresolutionsatelliteimages
AT huaguohuang combining3dradiativetransfermodelandconvolutionalneuralnetworktoaccuratelyestimateforestcanopycoverfromveryhighresolutionsatelliteimages
AT linyuanli combining3dradiativetransfermodelandconvolutionalneuralnetworktoaccuratelyestimateforestcanopycoverfromveryhighresolutionsatelliteimages
_version_ 1718425234928828416