Considering the classification of high spatial resolution remote sensing imagery, this paper presents a novel classification method for such imagery using deep neural networks. Deep learning methods, such as a fully convolutional network (FCN) model, achieve state-of-the-art performance in natural image semantic segmentation when provided with large-scale datasets and respective labels. To use data efficiently in the training stage, we first pre-segment training images and their labels into small patches as supplements of training data using graph-based segmentation and the selective search method. Subsequently, FCN with atrous convolution is used to perform pixel-wise classification. In the testing stage, post-processing with fully connected conditional random fields (CRFs) is used to refine results. Extensive experiments based on the Vaihingen dataset demonstrate that our method performs better than the reference state-of-the-art networks when applied to high-resolution remote sensing imagery classification.
Remote sensing (RS) images play a significant role in disaster emergency response. Web2.0 changes the way data are created, making it possible for the public to participate in scientific issues. In this paper, an experiment is designed to evaluate the reliability of crowdsourcing buildings collapse assessment in the early time after an earthquake based on aerial remote sensing image. The procedure of RS data pre-processing and crowdsourcing data collection is presented. A probabilistic model including maximum likelihood estimation (MLE), Bayes’ theorem and expectation-maximization (EM) algorithm are applied to quantitatively estimate the individual error-rate and “ground truth” according to multiple participants’ assessment results. An experimental area of Yushu earthquake is provided to present the results contributed by participants. Following the results, some discussion is provided regarding accuracy and variation among participants. The features of buildings labeled as the same damage type are found highly consistent. This suggests that the building damage assessment contributed by crowdsourcing can be treated as reliable samples. This study shows potential for a rapid building collapse assessment through crowdsourcing and quantitatively inferring “ground truth” according to crowdsourcing data in the early time after the earthquake based on aerial remote sensing image.
Considering the complexity of landslide hazards, their manual investigation lacks efficiency and is time-consuming, especially in high-altitude plateau areas. Therefore, extracting landslide information using remote sensing technology has great advantages. In this study, comprehensive research was carried out on the landslide features of high-resolution remote sensing images on the Mangkam dataset. Based on the idea of feature-driven classification, the landslide extraction model of a fully convolutional spectral–topographic fusion network (FSTF-Net) based on a deep convolutional neural network of multi-source data fusion is proposed, which takes into account the topographic factor (slope and aspect) and the normalized difference vegetation index (NDVI) as multi-source data input by which to train the model. In this paper, a high-resolution remote sensing image classification method based on a fully convolutional network was used to extract the landslide information, thereby realizing the accurate extraction of the landslide and surrounding ground-object information. With Mangkam County in the southeast of the Qinghai–Tibet Plateau China as the study area, the proposed method was evaluated based on the high-precision digital elevation model (DEM) generated from stereoscopic images of Resources Satellite-3 and multi-source high-resolution remote sensing image data (Beijing-2, Worldview-3, and SuperView-1). Results show that our method had a landslide detection precision of 0.85 and an overall classification accuracy of 0.89. Compared with the latest DeepLab_v3+, our model increases the landslide detection precision by 5%. Thus, the proposed FSTF-Net model has high reliability and robustness.
Rapid urbanization and economic development have led to the development of heavy industry and structural re-equalization in mainland China. This has resulted in scattered and disorderly layouts becoming prominent in the region. Furthermore, economic development has exacerbated pressures on regional resources and the environment and has threatened sustainable and coordinated development in the region. The NASA Land Science Investigator Processing System (Land-SIPS) Visible Infrared Imaging Radiometer (VIIRS) 375-m active fire product (VNP14IMG) was selected from the Fire Information for Resource Management System (FIRMS) to study the spatiotemporal patterns of heavy industry development. Furthermore, we employed an improved adaptive K-means algorithm to realize the spatial segmentation of long-order VNP14IMG and constructed heat source objects. Lastly, we used a threshold recognition model to identify heavy industry objects from normal heat source objects. Results suggest that the method is an accurate and effective way to monitor heat sources generated from heavy industry. Moreover, some conclusions about heavy industrial heat source distribution in mainland China at different scales were obtained. Those can be beneficial for policy-makers and heavy industry regulation.
Class imbalance is a key issue for the application of deep learning for remote sensing image classification because a model generated by imbalanced samples training has low classification accuracy for minority classes. In this study, an accurate classification approach using the multistage sampling method and deep neural networks was proposed to classify imbalanced data. We first balance samples by multistage sampling to obtain the training sets. Then, a state-of-the-art model is adopted by combining the advantages of atrous spatial pyramid pooling (ASPP) and Encoder-Decoder for pixel-wise classification, which are two different types of fully convolutional networks (FCNs) that can obtain contextual information of multiple levels in the Encoder stage. The details and spatial dimensions of targets are restored using such information during the Decoder stage. We employ four deep learning-based classification algorithms (basic FCN, FCN-8S, ASPP, and Encoder-Decoder with ASPP of our approach) on multistage training sets (original, MUS1, and MUS2) of WorldView-3 images in southeastern Qinghai-Tibet Plateau and GF-2 images in northeastern Beijing for comparison. The experiments show that, compared with existing sets (original, MUS1, and identical) and existing method (cost weighting), the MUS2 training set of multistage sampling significantly enhance the classification performance for minority classes. Our approach shows distinct advantages for imbalanced data.
Anthropogenic carbon dioxide (CO2) emissions from bottom-up inventories have high uncertainties due to the usage of proxy data in creating these inventories. To evaluate bottom-up inventories, satellite observations of atmospheric CO2 with continuously improved accuracies have shown great potential. In this study, we evaluate the consistency and uncertainty of four gridded CO2 emission inventories, including CHRED, PKU, ODIAC, and EDGAR that have been commonly used to study emissions in China, using GOSAT and OCO-2 satellite observations of atmospheric column-averaged dry-air mole fraction of CO2 (XCO2). The evaluation is carried out using two data-driven approaches: (1) quantifying the correlations of the four inventories with XCO2 anomalies derived from the satellite observations; (2) comparing emission inventories with emissions predicted by a machine learning-based model that considers the nonlinearity between emissions and XCO2. The model is trained using long-term datasets of XCO2 and emission inventories from 2010 to 2019. The result shows that the inconsistencies among these four emission inventories are significant, especially in areas of high emissions associated with large XCO2 values. In particular, EDGAR shows a larger difference to CHRED over super-emitting sources in China. The differences for ODIAC and EDGAR, when compared with the machine learning-based model, are higher in Asia than those in the USA and Europe. The predicted emissions in China are generally lower than the inventories, especially in megacities. The biases depend on the magnitude of inventory emissions with strong positive correlations with emissions (R2 is larger than 0.8). On the contrary, the predicted emissions in the USA are slightly higher than the inventories and the biases tend to be random (R2 is from 0.01 to 0.5). These results indicate that the uncertainties of gridded emission inventories of ODIAC and EDGAR are higher in Asian countries than those in European and the USA. This study demonstrates that the top-down approach using satellite observations could be applied to quantify the uncertainty of emission inventories and therefore improve the accuracy in spatially and temporally attributing national/regional totals inventories.