Physics-informed neural networks(PINNs)are promising to replace conventional mesh-based partial tial differen-equation(PDE)solvers by offering more accurate and flexible PDE solutions.However,PINNs are hampered by the...Physics-informed neural networks(PINNs)are promising to replace conventional mesh-based partial tial differen-equation(PDE)solvers by offering more accurate and flexible PDE solutions.However,PINNs are hampered by the relatively slow convergence and the need to perform additional,potentially expensive training for new PDE parameters.To solve this limitation,we introduce LatentPINN,a framework that utilizes latent representations of the PDE parameters as additional(to the coordinates)inputs into PINNs and allows for training over the distribution of these parameters.Motivated by the recent progress on generative models,we promote using latent diffusion models to learn compressed latent representations of the distribution of PDE parameters as they act as input parameters for NN functional solutions.We use a two-stage training scheme in which,in the first stage,we learn the latent representations for the distribution of PDE parameters.In the second stage,we train a physics-informed neural network over inputs given by randomly drawn samples from the coordinate space within the solution domain and samples from the learned latent representation of the PDE parameters.Considering their importance in capturing evolving interfaces and fronts in various fields,we test the approach on a class of level set equations given,for example,by the nonlinear Eikonal equation.We share results corresponding to three Eikonal parameters(velocity models)sets.The proposed method performs well on new phase velocity models without the need for any additional training.展开更多
Infrared and visible image fusion technology integrates the thermal radiation information of infrared images with the texture details of visible images to generate more informative fused images.However,existing method...Infrared and visible image fusion technology integrates the thermal radiation information of infrared images with the texture details of visible images to generate more informative fused images.However,existing methods often fail to distinguish salient objects from background regions,leading to detail suppression in salient regions due to global fusion strategies.This study presents a mask-guided latent low-rank representation fusion method to address this issue.First,the GrabCut algorithm is employed to extract a saliency mask,distinguishing salient regions from background regions.Then,latent low-rank representation(LatLRR)is applied to extract deep image features,enhancing key information extraction.In the fusion stage,a weighted fusion strategy strengthens infrared thermal information and visible texture details in salient regions,while an average fusion strategy improves background smoothness and stability.Experimental results on the TNO dataset demonstrate that the proposed method achieves superior performance in SPI,MI,Qabf,PSNR,and EN metrics,effectively preserving salient target details while maintaining balanced background information.Compared to state-of-the-art fusion methods,our approach achieves more stable and visually consistent fusion results.The fusion code is available on GitHub at:https://github.com/joyzhen1/Image(accessed on 15 January 2025).展开更多
Medical image fusion has been developed as an efficient assistive technology in various clinical applications such as medical diagnosis and treatment planning.Aiming at the problem of insufficient protection of image ...Medical image fusion has been developed as an efficient assistive technology in various clinical applications such as medical diagnosis and treatment planning.Aiming at the problem of insufficient protection of image contour and detail information by traditional image fusion methods,a new multimodal medical image fusion method is proposed.This method first uses non-subsampled shearlet transform to decompose the source image to obtain high and low frequency subband coefficients,then uses the latent low rank representation algorithm to fuse the low frequency subband coefficients,and applies the improved PAPCNN algorithm to fuse the high frequency subband coefficients.Finally,based on the automatic setting of parameters,the optimization method configuration of the time decay factorαe is carried out.The experimental results show that the proposed method solves the problems of difficult parameter setting and insufficient detail protection ability in traditional PCNN algorithm fusion images,and at the same time,it has achieved great improvement in visual quality and objective evaluation indicators.展开更多
Multimodal medical image fusion can help physicians provide more accurate treatment plans for patients, as unimodal images provide limited valid information. To address the insufficient ability of traditional medical ...Multimodal medical image fusion can help physicians provide more accurate treatment plans for patients, as unimodal images provide limited valid information. To address the insufficient ability of traditional medical image fusion solutions to protect image details and significant information, a new multimodality medical image fusion method(NSST-PAPCNNLatLRR) is proposed in this paper. Firstly, the high and low-frequency sub-band coefficients are obtained by decomposing the source image using NSST. Then, the latent low-rank representation algorithm is used to process the low-frequency sub-band coefficients;An improved PAPCNN algorithm is also proposed for the fusion of high-frequency sub-band coefficients. The improved PAPCNN model was based on the automatic setting of the parameters, and the optimal method was configured for the time decay factor αe. The experimental results show that, in comparison with the five mainstream fusion algorithms, the new algorithm has significantly improved the visual effect over the comparison algorithm,enhanced the ability to characterize important information in images, and further improved the ability to protect the detailed information;the new algorithm has achieved at least four firsts in six objective indexes.展开更多
Extracting discriminative speaker-specific representations from speech signals and transforming them into fixed length vectors are key steps in speaker identification and verification systems.In this study,we propose ...Extracting discriminative speaker-specific representations from speech signals and transforming them into fixed length vectors are key steps in speaker identification and verification systems.In this study,we propose a latent discriminative representation learning method for speaker recognition.We mean that the learned representations in this study are not only discriminative but also relevant.Specifically,we introduce an additional speaker embedded lookup table to explore the relevance between different utterances from the same speaker.Moreover,a reconstruction constraint intended to learn a linear mapping matrix is introduced to make representation discriminative.Experimental results demonstrate that the proposed method outperforms state-of-the-art methods based on the Apollo dataset used in the Fearless Steps Challenge in INTERSPEECH2019 and the TIMIT dataset.展开更多
representation that can identify and isolate different potential variables hidden in the highdimensional observations.Disentangled representation learning can capture information about a single change factor and contr...representation that can identify and isolate different potential variables hidden in the highdimensional observations.Disentangled representation learning can capture information about a single change factor and control it by the corresponding potential subspace,providing a robust representation for complex changes in the data.In this paper,we first introduce and analyze the current status of research on disentangled representation and its causal mechanisms and summarize three crucial properties of disentangled representation.Then,disentangled representation learning algorithms are classified into four categories and outlined in terms of both mathematical description and applicability.Subsequently,the loss functions and objective evaluation metrics commonly used in existing work on disentangled representation are classified.Finally,the paper summarizes representative applications of disentangled representation learning in the field of remote sensing and discusses its future development.展开更多
Multimodal medical imaging of human pathological tissues provides comprehensive information to assist in clinical diagnosis.However,due to the high cost of imaging,physiological incompatibility,and the harmfulness of ...Multimodal medical imaging of human pathological tissues provides comprehensive information to assist in clinical diagnosis.However,due to the high cost of imaging,physiological incompatibility,and the harmfulness of radioactive tracers,multimodal medical image data remains scarce.Currently,cross-modal medical synthesis methods can generate desired modal images from existing modal images.However,most existing methods are limited to specific domains.This paper proposes an Adaptive Domain Medical Image Synthesis Method based on Generative Adversarial Networks(ADGAN)to address this issue.ADGAN achieves multidirectional medical image synthesis and ensures pathological consistency by constructing a single generator to learn the latent shared representation of multiple domains.The generator employs dense connections in shallow layers to preserve edge details and incorporates auxiliary information in deep layers to retain pathological features.Additionally,spectral normalization is introduced into the discriminator to control discriminative performance and indirectly enhance the image synthesis ability of the generator.Theoretically,it can be proved that the proposed method can be trained quickly,and spectral normalization contributes to adaptive and multidirectional synthesis.In practice,comparing with recent state-of-the-art methods,ADGAN achieves average increments of 4.7%SSIM,6.7%MSIM,7.3%PSNR,and 9.2%VIF.展开更多
基金King Abdullah University of Science and Technol-ogy(KAUST)for supporting this research and the Seismic Wave Anal-ysis group for the supportive and encouraging environment.
文摘Physics-informed neural networks(PINNs)are promising to replace conventional mesh-based partial tial differen-equation(PDE)solvers by offering more accurate and flexible PDE solutions.However,PINNs are hampered by the relatively slow convergence and the need to perform additional,potentially expensive training for new PDE parameters.To solve this limitation,we introduce LatentPINN,a framework that utilizes latent representations of the PDE parameters as additional(to the coordinates)inputs into PINNs and allows for training over the distribution of these parameters.Motivated by the recent progress on generative models,we promote using latent diffusion models to learn compressed latent representations of the distribution of PDE parameters as they act as input parameters for NN functional solutions.We use a two-stage training scheme in which,in the first stage,we learn the latent representations for the distribution of PDE parameters.In the second stage,we train a physics-informed neural network over inputs given by randomly drawn samples from the coordinate space within the solution domain and samples from the learned latent representation of the PDE parameters.Considering their importance in capturing evolving interfaces and fronts in various fields,we test the approach on a class of level set equations given,for example,by the nonlinear Eikonal equation.We share results corresponding to three Eikonal parameters(velocity models)sets.The proposed method performs well on new phase velocity models without the need for any additional training.
基金supported by Universiti Teknologi MARA through UiTM MyRA Research Grant,600-RMC 5/3/GPM(053/2022).
文摘Infrared and visible image fusion technology integrates the thermal radiation information of infrared images with the texture details of visible images to generate more informative fused images.However,existing methods often fail to distinguish salient objects from background regions,leading to detail suppression in salient regions due to global fusion strategies.This study presents a mask-guided latent low-rank representation fusion method to address this issue.First,the GrabCut algorithm is employed to extract a saliency mask,distinguishing salient regions from background regions.Then,latent low-rank representation(LatLRR)is applied to extract deep image features,enhancing key information extraction.In the fusion stage,a weighted fusion strategy strengthens infrared thermal information and visible texture details in salient regions,while an average fusion strategy improves background smoothness and stability.Experimental results on the TNO dataset demonstrate that the proposed method achieves superior performance in SPI,MI,Qabf,PSNR,and EN metrics,effectively preserving salient target details while maintaining balanced background information.Compared to state-of-the-art fusion methods,our approach achieves more stable and visually consistent fusion results.The fusion code is available on GitHub at:https://github.com/joyzhen1/Image(accessed on 15 January 2025).
文摘Medical image fusion has been developed as an efficient assistive technology in various clinical applications such as medical diagnosis and treatment planning.Aiming at the problem of insufficient protection of image contour and detail information by traditional image fusion methods,a new multimodal medical image fusion method is proposed.This method first uses non-subsampled shearlet transform to decompose the source image to obtain high and low frequency subband coefficients,then uses the latent low rank representation algorithm to fuse the low frequency subband coefficients,and applies the improved PAPCNN algorithm to fuse the high frequency subband coefficients.Finally,based on the automatic setting of parameters,the optimization method configuration of the time decay factorαe is carried out.The experimental results show that the proposed method solves the problems of difficult parameter setting and insufficient detail protection ability in traditional PCNN algorithm fusion images,and at the same time,it has achieved great improvement in visual quality and objective evaluation indicators.
基金funded by the National Natural Science Foundation of China,grant number 61302188.
文摘Multimodal medical image fusion can help physicians provide more accurate treatment plans for patients, as unimodal images provide limited valid information. To address the insufficient ability of traditional medical image fusion solutions to protect image details and significant information, a new multimodality medical image fusion method(NSST-PAPCNNLatLRR) is proposed in this paper. Firstly, the high and low-frequency sub-band coefficients are obtained by decomposing the source image using NSST. Then, the latent low-rank representation algorithm is used to process the low-frequency sub-band coefficients;An improved PAPCNN algorithm is also proposed for the fusion of high-frequency sub-band coefficients. The improved PAPCNN model was based on the automatic setting of the parameters, and the optimal method was configured for the time decay factor αe. The experimental results show that, in comparison with the five mainstream fusion algorithms, the new algorithm has significantly improved the visual effect over the comparison algorithm,enhanced the ability to characterize important information in images, and further improved the ability to protect the detailed information;the new algorithm has achieved at least four firsts in six objective indexes.
基金Project supported by the National Natural Science Foundation of China(Nos.U1836220 and 61672267)the Qing Lan Talent Program of Jiangsu Province,Chinathe Jiangsu Province Key Research and Development Plan(Industry Foresight and Key Core Technology)(No.BE2020036)。
文摘Extracting discriminative speaker-specific representations from speech signals and transforming them into fixed length vectors are key steps in speaker identification and verification systems.In this study,we propose a latent discriminative representation learning method for speaker recognition.We mean that the learned representations in this study are not only discriminative but also relevant.Specifically,we introduce an additional speaker embedded lookup table to explore the relevance between different utterances from the same speaker.Moreover,a reconstruction constraint intended to learn a linear mapping matrix is introduced to make representation discriminative.Experimental results demonstrate that the proposed method outperforms state-of-the-art methods based on the Apollo dataset used in the Fearless Steps Challenge in INTERSPEECH2019 and the TIMIT dataset.
基金supported by the National Natural Science Foundation of China(Nos.61825103,62202349)the Natural Science Foundation of Hubei Province(Nos.2022CFB352,2020CFA001)the Key Research&Development of Hubei Province(No.2020BIB006).
文摘representation that can identify and isolate different potential variables hidden in the highdimensional observations.Disentangled representation learning can capture information about a single change factor and control it by the corresponding potential subspace,providing a robust representation for complex changes in the data.In this paper,we first introduce and analyze the current status of research on disentangled representation and its causal mechanisms and summarize three crucial properties of disentangled representation.Then,disentangled representation learning algorithms are classified into four categories and outlined in terms of both mathematical description and applicability.Subsequently,the loss functions and objective evaluation metrics commonly used in existing work on disentangled representation are classified.Finally,the paper summarizes representative applications of disentangled representation learning in the field of remote sensing and discusses its future development.
基金supported by the National Natural Science Foundation of China(Nos.62176217 and 62206224)the Innovation Team Funds of China West Normal University(No.KCXTD2022-3)+2 种基金the Postdoctoral Science Foundation of China(No.2023M732428)the Natural Science Foundation of Sichuan Province(No.2022NSFSC0866)the Doctoral Research Innovation Project(No.21E025).
文摘Multimodal medical imaging of human pathological tissues provides comprehensive information to assist in clinical diagnosis.However,due to the high cost of imaging,physiological incompatibility,and the harmfulness of radioactive tracers,multimodal medical image data remains scarce.Currently,cross-modal medical synthesis methods can generate desired modal images from existing modal images.However,most existing methods are limited to specific domains.This paper proposes an Adaptive Domain Medical Image Synthesis Method based on Generative Adversarial Networks(ADGAN)to address this issue.ADGAN achieves multidirectional medical image synthesis and ensures pathological consistency by constructing a single generator to learn the latent shared representation of multiple domains.The generator employs dense connections in shallow layers to preserve edge details and incorporates auxiliary information in deep layers to retain pathological features.Additionally,spectral normalization is introduced into the discriminator to control discriminative performance and indirectly enhance the image synthesis ability of the generator.Theoretically,it can be proved that the proposed method can be trained quickly,and spectral normalization contributes to adaptive and multidirectional synthesis.In practice,comparing with recent state-of-the-art methods,ADGAN achieves average increments of 4.7%SSIM,6.7%MSIM,7.3%PSNR,and 9.2%VIF.
