Search

This paper proposes an end-to-end neural network model that fully utilizes the characteristic of uneven fog distribution to estimate visibility in fog images. Firstly, we transform the original single labels into discrete label distributions and introduce discrete label distribution learning on top of the existing classification networks to learn the difference in visibility information among different regions of an image. Then, we employ the bilinear attention pooling module to find the farthest visible region of fog in the image, which is incorporated into an attention-based branch. Finally, we conduct a cascaded fusion of the features extracted from the attention-based branch and the base branch. Extensive experimental results on a real highway dataset and a publicly available synthetic road dataset confirm the effectiveness of the proposed method, which has low annotation requirements, good robustness, and broad application space. [ABSTRACT FROM AUTHOR]

A sun photometer is a type of photometer that points at the sun, and it has been playing an increasingly important role in characterizing aerosols across the world. As long as the solar photometer is accurately calibrated, the optical thickness of the aerosol can be obtained from the measured value of this device. When the calibration of a single instrument is not accurate, the inversion quantity varies greatly. The calibration constant of the sun photometer changes during its use process; thus, calibrations are frequently needed in order to ensure the accuracy of the measured value. The calibration constant of the solar photometer is usually determined using the Langley method. Internationally, AERONET has two Langley calibration stations: the Mauna Loa observatory in the United States and the Izaña observatory in Spain. So far, the International Comparison and Calibration System has been established in Beijing, similar to AERONET at GSFC, but the Langley calibration system has not yet been established. Therefore, it is necessary to select a suitable calibration station in China. This paper studies the requirements of the calibration station using the Langley method. We used long-term records of satellite-derived measurements and survey data belonging to the aerosol optical thickness data of SNPP/VIIRS, CERES, MERRA-2, etc., in order to gain a better understanding of whether these stations are suitable for calibration. From the existing astronomical observation stations, meteorological stations, and the Sun–Sky Radiometer Observation Network (SONET) observation stations in China, the qualified stations were selected. According to the statistical data from the Ali observatory, the monthly average of clear sky is 20.21 days, and it is always greater than 15 days. The monthly average of aerosol is not more than 0.15 and is less than 0.3. We believe that the atmosphere above the Ali observatory is stable, and the results show that the Ali observatory has excellent weather conditions. This study can provide a selection of calibration sites for solar photometer calibrations in China that may need to be further characterized and evaluated, and at the same time provide a method to exclude unsuitable calibration sites. [ABSTRACT FROM AUTHOR]

This paper proposes a novel end-to-end pipeline that uses the ordinal information and relative relation of images for visibility estimation (VISOR-NET). By encoding ordinal information into a set of relatively ordered image pairs, VISOR-NET can learn a global ranking function effectively. Due to the lack of real scenes or continuous labels in public foggy datasets, we collect a large-scale dataset that we term Foggy Highway Visibility Images (FHVI), which are taken from real surveillance scenes, and synthesize an INDoor Foggy images dataset (INDF) with continuous annotation. This work measures the estimation effectiveness on two public datasets and our FHVI dataset as a classification task and then on the INDF dataset as a regression task. Comprehensive experiments with existing deep-learning methods demonstrate the performance of the proposed method in terms of estimation accuracy, the convergence rate, model stability, and data requirements. Moreover, this method can extend inter-level visibility estimation to intra-level visibility estimation and can realize approximate regression estimation under discrete-level labels. [ABSTRACT FROM AUTHOR]

As one kind of remote sensing image (RSI), Directional Polarimetric Camera (DPC) data are of great significance in atmospheric radiation transfer and climate feedback. The availability of DPC images is often hindered by clouds, and effective cloud detection is the premise of many applications. Conventional threshold-based cloud detection methods are limited in performance and generalization capability. In this paper, we propose an effective learning-based 3D multimodal fusion cloud detection network (MFCD-Net) model. The network is a three-input stream architecture with a 3D-Unet-like encoder-decoder structure to fuse the multiple modalities of reflectance image, polarization image Q, and polarization image U in DPC imagery, with consideration of the angle and spectral information. Furthermore, cross attention is utilized in fusing the polarization features into the spatial-angle-spectral features in the reflectance image to enhance the expression of the fused features. The dataset used in this paper is obtained from the DPC cloud product and the cloud mask product. The proposed MFCD-Net achieved excellent cloud detection performance, with a recognition accuracy of 95.74%, according to the results of the experiments. [ABSTRACT FROM AUTHOR]

Cloud detection is one of the critical tasks in remote sensing image preprocessing. Remote sensing images usually contain multi-dimensional information, which is not utilized entirely in existing deep learning methods. This paper proposes a novel cloud detection algorithm based on multi-scale input and dual-channel attention mechanisms. Firstly, we remodeled the original data to a multi-scale layout in terms of channels and bands. Then, we introduced the dual-channel attention mechanism into the existing semantic segmentation network, to focus on both band information and angle information based on the reconstructed multi-scale data. Finally, a multi-scale fusion strategy was introduced to combine band information and angle information simultaneously. Overall, in the experiments undertaken in this paper, the proposed method achieved a pixel accuracy of 92.66% and a category pixel accuracy of 92.51%. For cloud detection, the proposed method achieved a recall of 97.76% and an F1 of 95.06%. The intersection over union (IoU) of the proposed method was 89.63%. Both in terms of quantitative results and visual effects, the deep learning model we propose is superior to the existing semantic segmentation methods. [ABSTRACT FROM AUTHOR]

Precipitation nowcasting plays an important role in the early warning of disasters and many other aspects of people's lives. In this study, we address the problem of radar reflectivity image extrapolation, which has great significance for precipitation near‐range forecasting. In recent years, the related achievements of nowcasting indicate that deep learning‐based methods have been far ahead of traditional ones. However, most deep learning methods focus on spatial appearance but cannot characterize the motion information well. To solve this complex problem, in addition to completing pixel‐level predictions by McNet, we introduce FlowNet and optical flow loss into generative adversarial networks (GAN) to express motion information more effectively. The reflectance image changes significantly due to complex meteorological conditions. Only by adding motion features can the law of change be determined. Thus, we can retain more details in the image. To the best of our knowledge, this is the first time FlowNet has been combined with GAN to fulfil the task of precipitation prediction. Extensive experiments on the dataset provided by the Shenzhen Meteorological Bureau demonstrate that our network performs favourably against other state‐of‐the‐art methods, which presents great guiding significance for precipitation nowcasting and possesses broad application prospects.Precipitation prediction process based on the optical flow method. [ABSTRACT FROM AUTHOR]

In recent years, deep learning has been widely used in the classification of hyperspectral images and good results have been achieved. But it is easy to ignore the edge information of the image when using the spatial features of hyperspectral images to carry out the classification experiments. In order to make full use of the advantages of convolution neural network (CNN), we extract the spatial information with the method of minimum noise fraction (MNF) and the edge information by bilateral filter. The combination of the two kinds of information not only increases the useful information but also effectively removes part of the noise. The convolution neural network is used to extract features and classify for hyperspectral images on the basis of this fused information. In addition, this paper also uses another kind of edge-filtering method to amend the final classification results for a better accuracy. The proposed method was tested on three public available data sets: the University of Pavia, the Salinas, and the Indian Pines. The competitive results indicate that our approach can realize a classification of different ground targets with a very high accuracy. [ABSTRACT FROM AUTHOR]

Recent deep object detection methods neglect the intrinsic properties of the license plate, which limits the detection performance in unconstrained scenes. In this paper, we propose a two-stage deep learning-based method to locate license plates in unconstrained scenes, especially for special license plates such as fouling, occlusion, and so on. A deep network consisting of convolutional neural network (CNN) and recurrent neural network is designed. In the first stage, fine-scale proposals are detected according to the characteristics of the license plate characters, and CNN is used to extract the local features of characters. A vertical anchor mechanism is designed to jointly predict the position and confidence of each fix-width character. Furthermore, the sequential contexts of characters are modeled with the bi-directional long short-term memory, which greatly improves the locating rate of license plates in complex scenes. In the second stage, the whole license plate is obtained by connecting the fine-scale proposals. The experimental results show that the proposed method not only locates license plates of different countries accurately but also be robust to scenes of illumination variation, noise distortion, and blurry effects. The average precision reaches 97.11% on multi-country license plates, and the precision and recall reaches 99.10% and 98.68%, respectively, on Chinese license plate images. [ABSTRACT FROM AUTHOR]

Recently, single object tracking has demonstrated great success. However, due to various problems caused by fast motion, occlusion, and deformation, it is still intractable for traditional trackers to adapt to changes in object appearance. In this work, a dynamic memory network is proposed for visual tracking to handle the template matching process globally. More specifically, we first build a memory model, which consists of a memory feature fusion module and a memory bank. By the memory model, the network not only accepts the first frame as initial information but also memorizes the selective frames in the video sequence to provide rich time-domain information. Second, an innovative sampling strategy is adopted in the tracking process. By updating the template and guiding the selection of memory frames, our model can output higher-quality features. In addition, a spatial-channel fused attention module that effectively improves the representational capability and discriminability of the model is introduced. Our proposed method obtains compelling results on six challenging tracking benchmarks, including the OTB100, VOT2019, UAV123, NFS, GOT-10k, and LaSOT datasets. Extensive experiments demonstrate that our approach shows satisfactory robustness and leading application potential in real-time speed. [ABSTRACT FROM AUTHOR]

We use two cloud screening methods—the clustering method and the multiplet method—to process the measurements of a sun photometer from March 2020 to April 2021 in Shouxian. The aerosol optical depth (AOD) and Angström parameters α and β are retrieved; variation characteristics and single scattering albedo are studied. The results show that: (1) The fitting coefficient of AOD retrieved by the two methods is 0.921, and the changing trend is consistent. The clustering method has fewer effective data points and days, reducing the overall average of AOD by 0.0542 (500 nm). (2) Diurnal variation of AOD can be divided into flat type, convex type, and concave type. Concave type and convex type occurred the most frequently, whereas flat type the least. (3) During observation, the overall average of AOD is 0.48, which is relatively high. Among them, AOD had a winter maximum (0.70), autumn and spring next (0.54 and 0.40), and a summer minimum (0.26). The variation trend of AOD and β is highly consistent, and the monthly mean of α is between 0.69 and 1.61, concerning mainly continental and urban aerosols. (4) Compared with others, the single scattering albedo in Shouxian is higher, reflecting strong scattering and weak aerosol absorption. [ABSTRACT FROM AUTHOR]

Background: Plantar pressure image (PPI) recorded in high spatial and temporal resolution is very useful in clinical gait analysis. For functional analysis of PPI, image registration is often performed to maximally correlate source image with a template image. Previous methods estimate the registration parameters by iteratively optimizing different objective functions. These methods are often computational expensive to achieve satisfactory registration accuracy.Research Question: Can we develop a single PPI registration technique that performs more rapidly than previous methods, and that also maintains adequate PPI correspondence as defined by various (dis)similarity metrics?Methods: A cascaded convolutional neural network (CNN) was proposed for the registration of PPIs. Our model was trained to learn a regression from the difference between the template and misaligned images to the registration parameters. The registration performance was evaluated by three different metrics, i.e. the mean squared error (MSE), the exclusive or (XOR), and the mutual information (MI). For comparison, four previous methods were also implemented. These included the principal axes (PA) method, the center of pressure trajectory (COP) method, the MSE method, and the XOR method.Results: Experimental results on a dataset with 71 PPI template-source pairs showed that the proposed CNN-based method could obtain comparable registration accuracy to the MSE and XOR method. With regards to the registration speed, registration durations (mean ± sd in seconds) per image pair were: MSE (30.584 ± 2.171), XOR (24.245 ± 1.596), PA (0.016 ± 0.003), COP (25.614 ± 0.341), and the proposed model (0.054 ± 0.007).Significance: Our findings indicate that the proposed registration approach can achieve high accuracy but less computational time. Thus, it is more practical to utilize our pre-trained CNN-based model to develop near-real time applications for plantar pressure images registration. [ABSTRACT FROM AUTHOR]