Radar Echo Spatiotemporal Sequence Prediction Using an Improved ConvGRU Deep Learning Model

He, Wei; Xiong, Taisong; Wang, Hao; He, Jinhai; Ren, Xinyue; Yan, Yilin; Tan, Linyin

doi:10.3390/atmos13010088

Cited by 13 publications

(8 citation statements)

References 23 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…ConvGRU uses convolution rather than dot product operation in GRU, enabling the model to more closely match the contextual information 10 , 11 . He et al 12 . used ConvGRU to capture the spatiotemporal features of radar echo sequences by performing several rounds of convolution-based gated processing on the input data x and the previous output state hpre to update x and hpre.…”

Section: Related Workmentioning

confidence: 99%

“…ConvGRU uses convolution rather than dot product operation in GRU, enabling the model to more closely match the contextual information. 10,11 He et al 12 used ConvGRU to capture the spatiotemporal features of radar echo sequences by performing several rounds of convolution-based gated processing on the input data x and the previous output state h pre to update x and h pre . Yuan et al 13 introduced attention mechanism into ConvGRU to fuse the intrinsic semantic relationship of features, but multiple attention ConvGRUs stacked together makes the network structure more complicated.…”

Section: Convgrumentioning

confidence: 99%

See 1 more Smart Citation

Multiscale feature fusion method for lane line detection based on time series

et al. 2023

View full text Add to dashboard Cite

.In order to solve lane line detection in difficult traffic conditions, such as shadow occlusion, signpost degradation, curves, and tunnels, numerous models have been proposed. However, most of the existing models conduct independent single-frame image detection, which makes it difficult to utilize the continuity of driving images and is ineffective in challenging scenes. To this end, we suggest a spatiotemporal information processing model for lane line recognition that enhances critical features. In order to properly learn the correlation between continuous images, we first employ a convolutional gated recurrent unit to process spatiotemporal driving information on the basis of U-Net. Second, the pyramid split attention (PSA) module is used to enhance or suppress the obtained feature expressions. Finally, the skip connection is used to fuse the features of different scales encoded by each stage with the features processed by PSA and gradually restore to the original image size. Experiments on the TuSimple dataset demonstrate that our model outperforms representative lane line detection networks in challenging driving scenes, with an F1-measure of up to 94.302%.

show abstract

Section: Related Workmentioning

confidence: 99%

Section: Convgrumentioning

confidence: 99%

Multiscale feature fusion method for lane line detection based on time series

et al. 2023

View full text Add to dashboard Cite

show abstract

“…Therefore, it is crucial for both individuals and governments to forecast approaching precipitation in advance and issue warnings, enabling appropriate response measures. Hence, enhancing the accuracy and timeliness of approaching precipitation forecasting bears great practical significance [1,2].…”

Section: Introductionmentioning

confidence: 99%

Residual Spatiotemporal Convolutional Neural Network Based on Multisource Fusion Data for Approaching Precipitation Forecasting

Zhang,

Wang,

Huang

et al. 2024

Atmosphere

View full text Add to dashboard Cite

Approaching precipitation forecast refers to the prediction of precipitation within a short time scale, which is usually regarded as a spatiotemporal sequence prediction problem based on radar echo maps. However, due to its reliance on single-image prediction, it lacks good capture of sudden severe convective events and physical constraints, which may lead to prediction ambiguities and issues such as false alarms and missed alarms. Therefore, this study dynamically combines meteorological elements from surface observations with upper-air reanalysis data to establish complex nonlinear relationships among meteorological variables based on multisource data. We design a Residual Spatiotemporal Convolutional Network (ResSTConvNet) specifically for this purpose. In this model, data fusion is achieved through the channel attention mechanism, which assigns weights to different channels. Feature extraction is conducted through simultaneous three-dimensional and two-dimensional convolution operations using a pure convolutional structure, allowing the learning of spatiotemporal feature information. Finally, feature fitting is accomplished through residual connections, enhancing the model’s predictive capability. Furthermore, we evaluate the performance of our model in 0–3 h forecasting. The results show that compared with baseline methods, this network exhibits significantly better performance in predicting heavy rainfall. Moreover, as the forecast lead time increases, the spatial features of the forecast results from our network are richer than those of other baseline models, leading to more accurate predictions of precipitation intensity and coverage area.

show abstract

“…Pixel-level prediction refers to the prediction of the pixels of the remote sensing image at the future time. Relevant research includes time series prediction of the remote sensing image based on the convolutional long short term memory network (ConvLSTM) ( Ma et al, 2021 ) and convolutional gated recurrent unit (ConvGRU) ( He et al, 2022 ), which uses zero padding in the convolution process, as a result, image edge information prediction is not very accurate; remote sensing image prediction based on generation adversarial network (GAN) and its variants such as Pix2Pix ( Li et al, 2019a ; Rüttgers et al, 2019 ), the adversarial learning method of this kind of methods can improve the prediction accuracy, but it is mainly based on the extraction of spatial features; remote sensing image prediction based on 3D-Unet and 3DPatchGAN’s 3D-Pix2Pix ( Bihlo, 2021 ) can simultaneously extract temporal and spatial features, however, gradient explosion or gradient disappearance is likely to occur during the training process due to the complex structure of the model and numerous parameters. Pixel-level prediction takes remote sensing images as input and output, and the prediction results are still images, which can reflect the future temporal and spatial distribution.…”

Section: Introductionmentioning

confidence: 99%

Remote sensing image analysis and prediction based on improved Pix2Pix model for water environment protection of smart cities

Wang

et al. 2023

PeerJ Computer Science

View full text Add to dashboard Cite

Background As an important part of smart cities, smart water environmental protection has become an important way to solve water environmental pollution problems. It is proposed in this article to develop a water quality remote sensing image analysis and prediction method based on the improved Pix2Pix (3D-GAN) model to overcome the problems associated with water environment prediction of smart cities based on remote sensing image data having low accuracy in predicting image information, as well as being difficult to train. Methods Firstly, due to inversion differences and weather conditions, water quality remote sensing images are not perfect, which leads to the creation of time series data that cannot be used directly in prediction modeling. Therefore, a method for preprocessing time series of remote sensing images has been proposed in this article. The original remote sensing image was unified by pixel substitution, the image was repaired by spatial weight matrix, and the time series data was supplemented by linear interpolation. Secondly, in order to enhance the ability of the prediction model to process spatial-temporal data and improve the prediction accuracy of remote sensing images, the convolutional gated recurrent unit network is concatenated with the U-net network as the generator of the improved Pix2Pix model. At the same time, the channel attention mechanism is introduced into the convolutional gated recurrent unit network to enhance the ability of extracting image time series information, and the residual structure is introduced into the downsampling of the U-net network to avoid gradient explosion or disappearance. After that, the remote sensing images of historical moments are superimposed on the channels as labels and sent to the discriminator for adversarial training. The improved Pix2Pix model no longer translates images, but can predict two dimensions of space and one dimension of time, so it is actually a 3D-GAN model. Third, remote sensing image inversion data of chlorophyll-a concentrations in the Taihu Lake basin are used to verify and predict the water environment at future moments. Results The results show that the mean value of structural similarity, peak signal-to-noise ratio, cosine similarity, and mutual information between the predicted value of the proposed method and the real remote sensing image is higher than that of existing methods, which indicates that the proposed method is effective in predicting water environment of smart cities.

show abstract

Radar Echo Spatiotemporal Sequence Prediction Using an Improved ConvGRU Deep Learning Model

Cited by 13 publications

References 23 publications

Multiscale feature fusion method for lane line detection based on time series

Multiscale feature fusion method for lane line detection based on time series

Residual Spatiotemporal Convolutional Neural Network Based on Multisource Fusion Data for Approaching Precipitation Forecasting

Remote sensing image analysis and prediction based on improved Pix2Pix model for water environment protection of smart cities

Contact Info

Product

Resources

About