Search

Cross-scene classification focuses on setting up an effective domain adaptation (DA) way to transfer the learnable knowledge from source to target domain, which can be reasonably achieved through the pseudo-label propagation procedure. However, it is hard to bridge the objective existing severe domain discrepancy between source and target domains, and thus, there are several unreliable pseudo-labels generated in target domain and involved into pseudo-label propagation procedure, which would lead to unreliable error accumulation to deteriorate the performance of cross-scene classification. Therefore, in this paper, a novel Masked Image Modeling Auxiliary Pseudo-Label Propagation called MIM-AP 2 with clustering central rectification strategy is proposed to improve the quality of pseudo-label propagation for cross-scene classification. First, in order to gracefully bridge the domain discrepancy and improve DA representation ability in-domain, a supervised class-token contrastive learning is designed to find the more consistent contextual clues to achieve knowledge transfer learning from source to target domain. At the same time, it is also incorporated with a self-supervised MIM mechanism according to a low random masking ratio to capture domain-specific information for improving the discriminability in-domain, which can lay a solid foundation for high-quality pseudo-label generation. Second, aiming to alleviate the impact of unreliable error accumulation, a clustering central rectification strategy is designed to adaptively update robustness clustering central representations to assist in rectifying unreliable pseudo-labels and learning a superior target domain specific classifier for cross-scene classification. Finally, extensive experiments are conducted on six cross-scene classification benchmarks, and the results are superior to other DA methods. The average accuracy reached 95.79%, which represents a 21.87% improvement over the baseline. This demonstrates that the proposed MIM-AP 2 can provide significantly improved performance. [ABSTRACT FROM AUTHOR]

Wind profiling within operating wind farms is important for both wind resource assessment and wind power prediction. With increasing wind turbine size, it is getting difficult to obtain wind profiles covering the turbine-affecting area due to the limited height of wind towers. In this study, a stepwise quality control and optimizing process for deriving high-quality near-surface wind profiles within wind farms is proposed. The method is based on the radial wind speed obtained by the Doppler Wind Lidar velocity-azimuth display (VAD) technique. The method is used to obtain the whole wind profile from ground level to the height affected by wind turbines within a utility-scale onshore wind farm, in northern China. Compared with the traditional carrier-to-noise ratio (CNR) filter-based quality control method, the proposed data processing method can significantly improve the accuracy of the derived wind. For a 10 m wind speed, an increase in coefficient of determination (R2) from 0.826 to 0.932, and a decrease in mean absolute error (MAE) from 1.231% to 0.927% are obtained; while for 70 m wind speed, R2 increased from 0.926 to 0.958, and MAE decreased from 1.023% to 0.771%. For wind direction, R2 increased from 0.978 to 0.992 at 10 m, and increased from 0.983 to 0.995 at 70 m. The optimized method also presents advantages in improving the accuracy of derived wind under complex wind environments, e.g., inside a wind farm, and increasing the data availability during clear nights. The proposed method could be used to derive wind profiles from below the minimum range of a vertically operating scanning Doppler Lidar to a height affected by wind turbines. Combined with Doppler beam-swinging (DBS) scanning data, the method could be used to obtain the complete wind profile in the boundary layer. These wind profiles could be further used to predict wind power and evaluate the climate and environmental effects of wind farms. [ABSTRACT FROM AUTHOR]

The escalating frequency and severity of extreme climate greatly impact the carbon dynamics of terrestrial ecosystems worldwide. To understand the multi-temporal response of net ecosystem productivity (NEP) to extreme climate, we investigated 11 temperature and precipitation extreme indices across different vegetation types in China. From 1981 to 2019, the results showed that NEP in China increased at a rate of 0.64 g·m−2·a−2. Extreme climate demonstrated a significant warming trend and a non-significant moistening trend; specifically, maximum daily minimum temperature (TNx) exhibited a significant increase at a rate of 0.34 °C/10 a, while maximum 5-day precipitation (Rx5day) showed an insignificant increase at a rate of 1.78 mm/10 a. NEP was significantly impacted by extreme temperature at the annual, seasonal, and monthly scales, but moderately impacted by extreme precipitation. Specifically, extreme temperature had the most significant effect on grassland, with minimal influence on cropland. In contrast, extreme precipitation had the most significant effect on forest, with minimal impact on cropland. Moreover, the lagged time for extreme precipitation was longer than that for extreme temperature. Extreme precipitation exhibited a corresponding lagged time of at least 2 months (p < 0.01), while extreme temperature exhibited a lagged time of at least 1 month (p < 0.01). The maximum lag time observed was 4 months (p < 0.01). Our findings provide valuable insights into the multi-temporal response of NEP to extreme climate in China and inform sustainable development practices in the region. [ABSTRACT FROM AUTHOR]

Remote sensing image classification (RSIC) is a classical and fundamental task in the intelligent interpretation of remote sensing imagery, which can provide unique labeling information for each acquired remote sensing image. Thanks to the potent global context information extraction ability of the multi-head self-attention (MSA) mechanism, visual transformer (ViT)-based architectures have shown excellent capability in natural scene image classification. However, in order to achieve powerful RSIC performance, it is insufficient to capture global spatial information alone. Specifically, for fine-grained target recognition tasks with high inter-class similarity, discriminative and effective local feature representations are key to correct classification. In addition, due to the lack of inductive biases, the powerful global spatial context representation capability of ViT requires lengthy training procedures and large-scale pre-training data volume. To solve the above problems, a hybrid architecture of convolution neural network (CNN) and ViT is proposed to improve the RSIC ability, called P 2 FEViT, which integrates plug-and-play CNN features with ViT. In this paper, the feature representation capabilities of CNN and ViT applying for RSIC are first analyzed. Second, aiming to integrate the advantages of CNN and ViT, a novel approach embedding CNN features into the ViT architecture is proposed, which can make the model synchronously capture and fuse global context and local multimodal information to further improve the classification capability of ViT. Third, based on the hybrid structure, only a simple cross-entropy loss is employed for model training. The model can also have rapid and comfortable convergence with relatively less training data than the original ViT. Finally, extensive experiments are conducted on the public and challenging remote sensing scene classification dataset of NWPU-RESISC45 (NWPU-R45) and the self-built fine-grained target classification dataset called BIT-AFGR50. The experimental results demonstrate that the proposed P 2 FEViT can effectively improve the feature description capability and obtain outstanding image classification performance, while significantly reducing the high dependence of ViT on large-scale pre-training data volume and accelerating the convergence speed. The code and self-built dataset will be released at our webpages. [ABSTRACT FROM AUTHOR]

Passive location technology has been greatly developed because of its low power consumption, long detection distance, good concealment, and strong anti-interference ability. Orthogonal frequency-division multiplexing (OFDM) is an efficient multi-carrier transmission technology, which is an important signal form of 5G communication. Researching passive locations for OFDM signals can realize the location of base stations, which is of great significance in the military. Space-borne passive location technology has a contradiction between wide coverage and high precision. Therefore, a single-satellite passive location algorithm for OFDM radiation sources based on the virtual synthetic aperture is proposed. The algorithm introduces virtual synthetic aperture technology, using antenna movement to accumulate data coherently over a long time period and synthesizing a long azimuth virtual aperture. In addition, it utilizes fast Fourier transform (FFT) to extract phase information at a specific frequency based on the multi-carrier modulation technology of the OFDM signal. Pilot technology of the communication system is used for phase compensation and noise reduction. Thus, the azimuth linear frequency modulation (LFM) signal containing the location information of the radiation source is obtained. The radiation source location can be obtained by range searching and azimuth focusing. Simulation results verify the effectiveness of the algorithm and show that the algorithm can realize high-precision and wide-coverage location for the OFDM radiation sources using a single antenna, turning the hardware structure into software to reduce the cost and complexity of the system. [ABSTRACT FROM AUTHOR]

Population distribution data with high spatiotemporal resolution are of significant value and fundamental to many application areas, such as public health, urban planning, environmental change, and disaster management. However, such data are still not widely available due to the limited knowledge of complex human activity patterns. The emergence of location-based service big data provides additional opportunities to solve this problem. In this study, we integrated ambient population data, nighttime light data, and building volume data; innovatively proposed a spatial downscaling framework for Baidu heat map data during work time and sleep time; and mapped the population distribution with high spatiotemporal resolution (i.e., hourly, 100 m) in Beijing. Finally, we validated the generated population distribution maps with high spatiotemporal resolution using the highest-quality validation data (i.e., mobile signaling data). The relevant results indicate that our proposed spatial downscaling framework for both work time and sleep time has high accuracy, that the distribution of the population in Beijing on a regular weekday shows "centripetal centralization at daytime, centrifugal dispersion at night" spatiotemporal variation characteristics, that the interaction between the purpose of residents' activities and the spatial functional differences leads to the spatiotemporal evolution of the population distribution, and that China's "surgical control and dynamic zero COVID-19" epidemic policy was strongly implemented. In addition, our proposed spatial downscaling framework can be transferred to other regions, which is of value for governmental emergency measures and for studies about human risks to environmental issues. [ABSTRACT FROM AUTHOR]

Currently, under supervised learning, a model pre-trained by a large-scale nature scene dataset and then fine-tuned on a few specific task labeling data is the paradigm that has dominated knowledge transfer learning. Unfortunately, due to different categories of imaging data and stiff challenges of data annotation, there is not a large enough and uniform remote sensing dataset to support large-scale pre-training in the remote sensing domain (RSD). Moreover, pre-training models on large-scale nature scene datasets by supervised learning and then directly fine-tuning on diverse downstream tasks seems to be a crude method, which is easily affected by inevitable incorrect labeling, severe domain gaps and task-aware discrepancies. Thus, in this paper, considering the self-supervised pre-training and powerful vision transformer (ViT) architecture, a concise and effective knowledge transfer learning strategy called ConSecutive Pre-Training (CSPT) is proposed based on the idea of not stopping pre-training in natural language processing (NLP), which can gradually bridge the domain gap and transfer large-scale data knowledge to any specific domain (e.g., from nature scene domain to RSD) In addition, the proposed CSPT also can release the huge potential of unlabeled data for task-aware model training. Finally, extensive experiments were carried out on twelve remote sensing datasets involving three types of downstream tasks (e.g., scene classification, object detection and land cover classification) and two types of imaging data (e.g., optical and synthetic aperture radar (SAR)). The results show that by utilizing the proposed CSPT for task-aware model training, almost all downstream tasks in the RSD can outperform the previous knowledge transfer learning strategies based on model pre-training without any expensive manually labeling and even surpass the state-of-the-art (SOTA) performance without any careful network architecture designing. [ABSTRACT FROM AUTHOR]

Vegetation in the terrestrial ecosystem, sensitive to climate change and human activities, exerts a crucial influence on the carbon cycles in land, ocean, and atmosphere. Discrimination between climate and human-induced vegetation dynamics is advocated but still limited, especially in coastal China, which is characterized by a developed economy, a large population, and high food production, but also by unprecedented climate change and warming. Taking coastal China as the research area, our study used the normalized difference vegetation index (NDVI) in growing seasons, as well as precipitation, temperature, and sunlight hours datasets, adopted residual trend analysis at pixel and regional scales in coastal China from 2000–2019 and aims to (1) delineate the patterns and processes of vegetation changes, and (2) separate the relative contributions of climate and human activities by adopting residual trend analysis. The results indicated that (1) coastal China experienced the most vegetation greening (83.04% of the whole region) and partial degradation (16.86% of the whole region) with significant spatial heterogeneity; (2) compared with climate change, human activities have a greater positive impact on NDVI, and the regions were mainly located in the north of the North China Plain and the south of southern China; (3) the relative contribution rates of climate change and human activities were detected to be 0–60% and 60–100%, respectively; (4) in the northern coastal areas, the improvement of cultivated land management greatly promoted the greening of vegetation and thus the increase of grain yield, while in southern coastal areas, afforestation and the restoration of degraded forest were responsible for vegetation restoration; and (5) similar results obtained by partial correlation between nighttime lights and NDVI indicated the reliability of the residual trend analysis. The linear relationships of precipitation, temperature, and radiation on NDVI may limit the accurate estimation of climate drivers on vegetation, and further ecosystem process-modeling approaches can be used to estimate the relative contribution of climate change and human activities. The findings in our research emphasized that the attribution for vegetation dynamics with heterogeneity can provide evidence for the designation of rational ecological conservation policies. [ABSTRACT FROM AUTHOR]

Due to the proliferation of precision agriculture, the obstacle of estimating evapotranspiration (ET) and its components from shadow pixels acquired from remote sensing technology should not be neglected. To accurately detect shaded soil and leaf pixels and quantify the implications of shadow pixels on ET inversion, a two-year field-scale observation was carried out in the growing season for a pinot noir vineyard. Based on high-resolution remote sensing sensors covering visible light, thermal infrared, and multispectral light, the supervised classification was applied to detect shadow pixels. Then, we innovatively combined the normalized difference vegetation index with the three-temperature model to quantify the proportion of plant transpiration (T) and soil evaporation (E) in the vineyard ecosystem. Finally, evaluated with the eddy covariance system, we clarified the implications of the shadow pixels on the ET estimation and the spatiotemporal patterns of ET in a vineyard system by considering where shadow pixels were presented. Results indicated that the shadow detection process significantly improved reliable assessment of ET and its components. (1) The shaded soil pixels misled the land cover classification, with the mean canopy cover ignoring shadows 1.68–1.70 times more often than that of shaded area removal; the estimation accuracy of ET can be improved by 4.59–6.82% after considering the effect of shaded soil pixels; and the accuracy can be improved by 0.28–0.89% after multispectral correction. (2) There was a 2 °C canopy temperature discrepancy between sunlit leaves and shaded leaves, meaning that the estimation accuracy of T can be improved by 1.38–7.16% after considering the effect of shaded canopy pixels. (3) Simultaneously, the characteristics showed that there was heterogeneity of ET in the vineyard spatially and that E and T fluxes accounted for 238.05 and 208.79 W·m−2, respectively; the diurnal variation represented a single-peak curve, with a mean of 0.26 mm/h. Our findings provide a better understanding of the influences of shadow pixels on ET estimation using remote sensing techniques. [ABSTRACT FROM AUTHOR]

One of the problems of optical remote sensing of crop above-ground biomass (AGB) is that vegetation indices (VIs) often saturate from the middle to late growth stages. This study focuses on combining VIs acquired by a consumer-grade multiple-spectral UAV and machine learning regression techniques to (i) determine the optimal time window for AGB estimation of winter wheat and to (ii) determine the optimal combination of multi-spectral VIs and regression algorithms. UAV-based multi-spectral data and manually measured AGB of winter wheat, under five nitrogen rates, were obtained from the jointing stage until 25 days after flowering in the growing season 2020/2021. Forty-four multi-spectral VIs were used in the linear regression (LR), partial least squares regression (PLSR), and random forest (RF) models in this study. Results of LR models showed that the heading stage was the most suitable stage for AGB prediction, with R2 values varying from 0.48 to 0.93. Three PLSR models based on different datasets performed differently in estimating AGB in the training dataset (R2 = 0.74~0.92, RMSE = 0.95~2.87 t/ha, MAE = 0.75~2.18 t/ha, and RPD = 2.00~3.67) and validation dataset (R2 = 0.50~0.75, RMSE = 1.56~2.57 t/ha, MAE = 1.44~2.05 t/ha, RPD = 1.45~1.89). Compared with PLSR models, the performance of the RF models was more stable in the prediction of AGB in the training dataset (R2 = 0.95~0.97, RMSE = 0.58~1.08 t/ha, MAE = 0.46~0.89 t/ha, and RPD = 3.95~6.35) and validation dataset (R2 = 0.83~0.93, RMSE = 0.93~2.34 t/ha, MAE = 0.72~2.01 t/ha, RPD = 1.36~3.79). Monitoring AGB prior to flowering was found to be more effective than post-flowering. Moreover, this study demonstrates that it is feasible to estimate AGB for multiple growth stages of winter wheat by combining the optimal VIs and PLSR and RF models, which overcomes the saturation problem of using individual VI-based linear regression models. [ABSTRACT FROM AUTHOR]

We present a GPU-based computation for simulating the synthetic aperture radar (SAR) image of the complex target. To be more realistic, we included the multiple scattering field and antenna pattern tracking in producing the SAR echo signal for both Stripmap and Spotlight modes. Of the signal chains, the computation of the backscattering field is the most computationally intensive. To resolve the issue, we implement a computation parallelization for SAR echo signal generation. By profiling, the overall processing was identified to find which is the heavy loading stage. To further accommodate the hardware structure, we made extensive modifications in the CUDA kernel function. As a result, the computation efficiency is much improved, with over 224 times the speed up. The computation complexity by comparing the CPU and GPU computations was provided. We validated the proposed simulation algorithm using canonical targets, including a perfectly electric conductor (PEC), dielectric spheres, and rotated/unrotated dihedral corner reflectors. Additionally, the targets can be a multi-layered dielectric coating or a layered medium. The latter case aimed to evaluate the polarimetric response quantitively. Then, we simulated a complex target with various poses relative to the SAR imaging geometry. We show that the simulated images have high fidelity in geometric and radiometric specifications. The decomposition of images from individual scattering bounce offers valuable exploitation of the scattering mechanisms responsible for imaging certain target features. [ABSTRACT FROM AUTHOR]

Deep-learning technologies, especially convolutional neural networks (CNNs), have achieved great success in building extraction from areal images. However, shape details are often lost during the down-sampling process, which results in discontinuous segmentation or inaccurate segmentation boundary. In order to compensate for the loss of shape information, two shape-related auxiliary tasks (i.e., boundary prediction and distance estimation) were jointly learned with building segmentation task in our proposed network. Meanwhile, two consistency constraint losses were designed based on the multi-task network to exploit the duality between the mask prediction and two shape-related information predictions. Specifically, an atrous spatial pyramid pooling (ASPP) module was appended to the top of the encoder of a U-shaped network to obtain multi-scale features. Based on the multi-scale features, one regression loss and two classification losses were used for predicting the distance-transform map, segmentation, and boundary. Two inter-task consistency-loss functions were constructed to ensure the consistency between distance maps and masks, and the consistency between masks and boundary maps. Experimental results on three public aerial image data sets showed that our method achieved superior performance over the recent state-of-the-art models. [ABSTRACT FROM AUTHOR]

Remote sensing images contain complex backgrounds and multi-scale objects, which pose a challenging task for scene classification. The performance is highly dependent on the capacity of the scene representation as well as the discriminability of the classifier. Although multiple models possess better properties than a single model on these aspects, the fusion strategy for these models is a key component to maximize the final accuracy. In this paper, we construct a novel dual-model architecture with a grouping-attention-fusion strategy to improve the performance of scene classification. Specifically, the model employs two different convolutional neural networks (CNNs) for feature extraction, where the grouping-attention-fusion strategy is used to fuse the features of the CNNs in a fine and multi-scale manner. In this way, the resultant feature representation of the scene is enhanced. Moreover, to address the issue of similar appearances between different scenes, we develop a loss function which encourages small intra-class diversities and large inter-class distances. Extensive experiments are conducted on four scene classification datasets include the UCM land-use dataset, the WHU-RS19 dataset, the AID dataset, and the OPTIMAL-31 dataset. The experimental results demonstrate the superiority of the proposed method in comparison with the state-of-the-arts. [ABSTRACT FROM AUTHOR]