Comparing Fully Deep Convolutional Neural Networks for Land Cover Classification with High-Spatial-Resolution Gaofen-2 Images

Han, Zemin; Dian, Yuanyong; Xia, Hao; Zhou, Jingjing; Jian, Yongfeng; Yao, Chonghuai; Li, Yuan

doi:10.3390/ijgi9080478

Cited by 32 publications

(17 citation statements)

References 28 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The designs of the ANNs result in the diversity of their advancements and limitations [30]. For example, the convolutional neural networks have the kernel to recognize the spatial distribution pattern of data; consequently, it has been implemented commonly for imagery classification [31][32][33], however, the existence of pooling layers may cause the loss of abundant information [34]. The recurrent neural network (RNN) is a "loop" structure that has a state to store the information from the previous steps; thus, it can deal with the time-series data that has time dependency.…”

Section: Introductionmentioning

confidence: 99%

Impact of Input Filtering and Architecture Selection Strategies on GRU Runoff Forecasting: A Case Study in the Wei River Basin, Shaanxi, China

Wang

Liu

Yue

et al. 2020

Water

View full text Add to dashboard Cite

A gated recurrent unit (GRU) network, which is a kind of artificial neural network (ANN), has been increasingly applied to runoff forecasting. However, knowledge about the impact of different input data filtering strategies and the implications of different architectures on the GRU runoff forecasting model’s performance is still insufficient. This study has selected the daily rainfall and runoff data from 2007 to 2014 in the Wei River basin in Shaanxi, China, and assessed six different scenarios to explore the patterns of that impact. In the scenarios, four manually-selected rainfall or runoff data combinations and principal component analysis (PCA) denoised input have been considered along with single directional and bi-directional GRU network architectures. The performance has been evaluated from the aspect of robustness to 48 various hypermeter combinations, also, optimized accuracy in one-day-ahead (T + 1) and two-day-ahead (T + 2) forecasting for the overall forecasting process and the flood peak forecasts. The results suggest that the rainfall data can enhance the robustness of the model, especially in T + 2 forecasting. Additionally, it slightly introduces noise and affects the optimized prediction accuracy in T + 1 forecasting, but significantly improves the accuracy in T + 2 forecasting. Though with relevance (R = 0.409~0.763, Grey correlation grade >0.99), the runoff data at the adjacent tributary has an adverse effect on the robustness, but can enhance the accuracy of the flood peak forecasts with a short lead time. The models with PCA denoised input has an equivalent, even better performance on the robustness and accuracy compared with the models with the well manually filtered data; though slightly reduces the time-step robustness, the bi-directional architecture can enhance the prediction accuracy. All the scenarios provide acceptable forecasting results (NSE of 0.927~0.951 for T + 1 forecasting and 0.745~0.836 for T + 2 forecasting) when the hyperparameters have already been optimized. Based on the results, recommendations have been provided for the construction of the GRU runoff forecasting model.

show abstract

Section: Introductionmentioning

confidence: 99%

Impact of Input Filtering and Architecture Selection Strategies on GRU Runoff Forecasting: A Case Study in the Wei River Basin, Shaanxi, China

Wang

Liu

Yue

et al. 2020

Water

View full text Add to dashboard Cite

show abstract

“…The white lines are the main roads in the study area obtained from openstreetmap.org based model are set by referring to Long et al (2015), while the RF part is formed by grid optimization. These models and parameters have been widely used in image semantic segmentation (Zhou et al, 2016), segmentation of street view images (Middel et al, 2019), remote sensing image classification (Han et al, 2020;Piramanayagam et al, 2016), public health (Zamani Joharestani et al, 2019), and proved to be effective. To better compare the CNN-based model, we refer to (Bulat & Tzimiropoulos, 2016;Simonyan & Zisserman, 2014) and other models to set parameters for the CNN's hyperparameters.…”

Section: Comparison Of Model Accuracy Based On Street View Datasetmentioning

confidence: 99%

A comparison of two deep-learning-based urban perception models: which one is better?

et al. 2021

View full text Add to dashboard Cite

Urban perception is a hot topic in current urban study and plays a positive role in urban planning and design. At present, there are two methods to calculate urban perception. 1) Using a model to learn image features directly automatically; 2) Coupling machine learning and feature extraction based on expert knowledge (e.g. object proportion) method. With two typical streets in Wuhan as the study area, video data were recorded and used as the model input. In this study, two representative methods are selected: 1) End to end convolution neural network (CNN-based model); 2) Based on full convolution neural network and random forest (FCN + RF-based model). By comparing the accuracy of two models, we analyze the adaptability of the model in different urban scenes. We also analyze the relationship between CNN-based model and urban function based on POI data and OSM data, and verify its interpretability. The results show that the CNN-based model is more accurate than FCN + RF-based model. Because the CNN-based model considers the topological characteristics of the ground objects, its perception results have a stronger nonlinear correlation with urban functions. In addition, we also find that the CNN-based model is more suitable for scenes with weak spatial heterogeneity (such as small and medium-sized urban environments), while the FCN + RF-based model is applicable to scenes with strong spatial heterogeneity (such as the downtown areas of China’s megacities). The results of this study can be used as a reference to provide decision support for urban perception model selection in urban planning.

show abstract

Section: Introductionmentioning

confidence: 99%

“…FCNNs and their variants have been used for end-to-end per-pixel image classification of high-resolution satellite images [29][30][31][32][33]. The number of classification classes varies with a maximum of 13 classes seen in [30]. For studies with more than 10 classes, high-resolution GaoFen-1, GaoFen-2, IKONOS, WV-2, and Quickbird satellite imagery were classified [29,30,32].…”

Section: Introductionmentioning

confidence: 99%

“…For studies with more than 10 classes, high-resolution GaoFen-1, GaoFen-2, IKONOS, WV-2, and Quickbird satellite imagery were classified [29,30,32]. However, these studies that use fully convolutional networks focus on the classification of land covers for an urban landscape [29,32] and agricultural landscapes [30] with a few focusing on ecological landscapes [34].The importance of high spatial resolution (3.7 m) SWIR bands for image classification, especially forest cover classification, has been unknown because 3.7 m SWIR bands have not been previously available for commercial or academic research. To the authors' best knowledge, FCNNs, that incorporate spatial features, have not been utilized for tropical forest cover mapping using raw WV-3 satellite imagery.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Forest Conservation with Deep Learning: A Deeper Understanding of Human Geography around the Betampona Nature Reserve, Madagascar

et al. 2021

View full text Add to dashboard Cite

Documenting the impacts of climate change and human activities on tropical rainforests is imperative for protecting tropical biodiversity and for better implementation of REDD+ and UN Sustainable Development Goals. Recent advances in very high-resolution satellite sensor systems (i.e., WorldView-3), computing power, and machine learning (ML) have provided improved mapping of fine-scale changes in the tropics. However, approaches so far focused on feature extraction or the extensive tuning of ML parameters, hindering the potential of ML in forest conservation mapping by not using textural information, which is found to be powerful for many applications. Additionally, the contribution of shortwave infrared (SWIR) bands in forest cover mapping is unknown. The objectives were to develop end-to-end mapping of the tropical forest using fully convolution neural networks (FCNNs) with WorldView-3 (WV-3) imagery and to evaluate human impact on the environment using the Betampona Nature Reserve (BNR) in Madagascar as the test site. FCNN (U-Net) using spatial/textural information was implemented and compared with feature-fed pixel-based methods including Support Vector Machine (SVM), Random Forest (RF), and Deep Neural Network (DNN). Results show that the FCNN model outperformed other models with an accuracy of 90.9%, while SVM, RF, and DNN provided accuracies of 88.6%, 84.8%, and 86.6%, respectively. When SWIR bands were excluded from the input data, FCNN provided superior performance over other methods with a 1.87% decrease in accuracy, while the accuracies of other models—SVM, RF, and DNN—decreased by 5.42%, 3.18%, and 8.55%, respectively. Spatial–temporal analysis showed a 0.7% increase in Evergreen Forest within the BNR and a 32% increase in tree cover within residential areas likely due to forest regeneration and conservation efforts. Other effects of conservation efforts are also discussed.

show abstract

Comparing Fully Deep Convolutional Neural Networks for Land Cover Classification with High-Spatial-Resolution Gaofen-2 Images

Cited by 32 publications

References 28 publications

Impact of Input Filtering and Architecture Selection Strategies on GRU Runoff Forecasting: A Case Study in the Wei River Basin, Shaanxi, China

Impact of Input Filtering and Architecture Selection Strategies on GRU Runoff Forecasting: A Case Study in the Wei River Basin, Shaanxi, China

A comparison of two deep-learning-based urban perception models: which one is better?

Forest Conservation with Deep Learning: A Deeper Understanding of Human Geography around the Betampona Nature Reserve, Madagascar

Contact Info

Product

Resources

About