Magnetotelluric (MT) forward modeling is fundamental for improving the accuracy and efficiency of MT inversion. Neural operators (NOs) have been effectively used for rapid MT forward modeling, demonstrating their promising performance in solving the MT forward modeling-related partial differential equations (PDEs). Particularly, they can obtain the electromagnetic field at arbitrary locations and frequencies. In these NOs, the projection layers have been dominated by multi-layer perceptrons (MLPs), which may potentially reduce the accuracy of solution due to they usually suffer from the disadvantages of MLPs, such as lack of interpretability, overfitting, and so on. Therefore, to improve the accuracy of MT forward modeling with NOs and explore the potential alternatives to MLPs, we propose a novel neural operator by extending the Fourier neural operator (FNO) with Kolmogorov-Arnold network (EFKAN). Within the EFKAN framework, the FNO serves as the branch network to calculate the apparent resistivity and phase from the resistivity model in the frequency domain. Meanwhile, the KAN acts as the trunk network to project the resistivity and phase, determined by the FNO, to the desired locations and frequencies. Experimental results demonstrate that the proposed method not only achieves higher accuracy in obtaining apparent resistivity and phase compared to the NO equipped with MLPs at the desired frequencies and locations but also outperforms traditional numerical methods in terms of computational speed.

High-resolution satellite remote sensing images are often compromised by cloud interference because of constraints,such as imaging conditions and sensing equipment.This phenomenon considerably degrades image quality and hinders intelligent interpretation.In recent years,deep learning has been extensively applied to cloud removal in remote sensing images because of its powerful feature learning capabilities.This study presents an attention-based generative adversarial network for cloud removal in remote sensing images.The network aims to address the challenge of preserving textures.Initially,a Fourier residual block is designed to integrate low-and high-frequency cloud images.Subsequently,attention mechanisms are employed for channel and spatial dimensions to capture the spatial distribution information of clouds that guides the network in effectively removing clouds.Then,the model is trained to minimize the joint loss function and ensure the coherence of generated images with real images in terms of content.Experimental results derived from the Remote Sensing Image Cloud Removing(RICE)dataset show that the proposed cloud removal method,which integrates attention mechanisms and generative adversarial networks,can effectively eradicate clouds while preserving textures.The following conclusions are drawn.(1)The proposed Fourier residual block plays a pivotal role in adeptly capturing the details of low-and high-frequency residual information present in cloud images.By merging these diverse features,the block bolsters the network's ability to extract rich features.(2)The attention mechanisms utilized extract features from spatial and channel dimensions.This extraction enables the model to discern the importance of various features across different dimensions.Moreover,the cloud information obtained through these attention mechanisms allows the model to adaptively reveal cloud changes,thereby enhancing the model's precision and efficiency in cloud removal.(3)Experimental results from the RICE dataset provide compelling evidence of the proposed method's effectiveness in terms of cloud removal and texture preservation.

Binary change detection (BCD) in remote sensing images is essential for tracking environmental and urban transformations, yet it grapples with challenges like scene complexity, multi-scale variations, and pseudo-changes due to illumination or seasonal shifts. Conventional convolutional neural networks(CNNs) are limited by fixed receptive fields. This limitation hinders their ability to effectively capture multi-scale variations and irregular shapes prevalent in remote sensing imagery. To overcome these constraints, we introduce DCNet that leverages deformable convolution for adaptive feature extraction. DCNet features a hierarchical encoder with deformable convolution and an intertemporal feature interaction module to mitigate pseudo-changes. By dynamically adjusting sampling locations, DCNet captures multi-scale and irregular changes, achieving state-of-the-art performance. Evaluations on the LEVIR-CD and WHU-CD datasets show DCNet surpassing state-of-the-art methods, with IoU improvements of 0.54% and 2.29%, respectively.

Semantic change detection of remote sensing images plays an important role in ecological environment,land use,and land cover monitoring.In recent years,deep learning-based change detection methods are the hotspot of remote sensing intelligent interpretation concern;however,the existing three-branch semantic change detection methods lack consistency of change branch and semantic branches,which leads to the self-contradiction of the bitemporal semantic change detection. To address this problem,this study proposes a remote sensing image semantic change detection algorithm based on Siamese CNN and Transformer.In the encoding stage,the Siamese ResNet34 network is first designed to extract the multiscale features of the image,and the difference enhancement module is embedded to increase the attention of the change information;then,the semantic tokenizer is used to map the feature map into compact semantic tokens,and the Transformer encoder is used to combine the bitemporal semantics and the change information to model the"semantic change"consistency.In the decoding stage,different fine-grained semantic information is fused by the Transformer decoder using a hopping connection to generate a refined semantic feature map.Finally,the result of bitemporal semantic change is obtained after upsampling recovery and mask multiplication. Experimental results on the remote sensing semantic change detection public dataset SECOND and Landsat-SCD show that the proposed algorithm can effectively focus on the change region,maintain the consistency between the change results and the semantic results,and achieve excellent evaluation indexes and visual effects. We can draw the following conclusions:(1)The proposed difference enhancement module can enhance the difference characteristics of bitemporal remote sensing images and improve the network's focus on change information.(2)The proposed bitemporal Transformer module maps the difference information and bitemporal semantic information into semantic tokens and fuses them to model jointly the"semantic change"information of the whole spatiotemporal domain in the token space,effectively modeling the long-range dependencies in the images and the bitemporal contextual correlations.The long-range dependency in the image is effectively modeled,and the bitemporal contextual relevance is modeled.The ReTNet network designed accordingly pays more attention to the change area and can accurately detect the change location and recognize the change element type of the bitemporal remote sensing image.

Deep learning is a powerful technique for semantic change detection (SCD) of bitemporal remote sensing images. In this work, we propose to improve SCD accuracy using deep learning with frequency feature enhancement (FFE). Specifically, we develop an FFE module that aims to enhance the performance of both binary change detection (BCD) and semantic segmentation, two main key components for obtaining high SCD accuracy, by integrating the Fourier transform and attention mechanisms. Experimental results on the SECOND and LandSat-SCD datasets demonstrate the effectiveness of the proposed method, and it achieves high resolution for change boundaries.

Objective Change detection in remote sensing imagery is a process that leverages remote sensing technology to compare and analyze images from the same geographical area but captured at different time intervals.This process mainly aims to identify changes on the Earth's surface.The main challenge in this process lies in the extraction of effective change features from a large volume of image data and subsequently mapping them onto pixel-level change labels for high-precision detection.Detection methods for changes in remote sensing imagery can be broadly divided into traditional and deep learning-based methods.Traditional methods primarily rely on image processing and pattern recognition techniques.How-ever,these methods often require manual selection of suitable features and thresholds,which can introduce a degree of sub-jectivity and limitations.By contrast,deep learning methods can automatically learn abstract and high-level change fea-tures from remote sensing images,thereby enabling end-to-end change detection.This approach notably enhances the accu-racy and efficiency of change detection.Among them,change detection models based on convolutional neural networks(CNNs)and Transformer architectures have shown remarkable performance.Models that utilize these mechanisms have demonstrated notable advancements in recent years due to the extensive research conducted by scholars worldwide.How-ever,for the currently effective models based on the Transformer architecture,the complexity of the model also increases as the detection accuracy of the model improves.Therefore,designing a change detection method with a small parameter size,low computational cost,and high detection accuracy is a pressing issue that must be urgently addressed in this field.Method This paper proposes a lightweight change detection method for remote sensing images based on feature interaction and fusion.The main idea of this method is to use EfficientNet B7 as a lightweight backbone network to extract deep-and low-level features from bi-temporal remote sensing images.Channel and pixel swap modules are introduced to enable the interaction and combination of bi-temporal features,enhancing the spatiotemporal feature fusion.Low-level skip-connections are employed to transfer the original image information to the upsampling phase,aiming to preserve additional edge and texture details and reduce artifact generation.A feature fusion group convolution module that reduces the compu-tational overhead and the number of parameters is designed to effectively fuse the deep-and low-level features obtained in the downsampling stage.Finally,the feature fusion group convolution and upsampling modules are used to fuse and recover the deep-and low-level features,and the pixel-level change detection map is generated.Result In this paper,experiments on two datasets are conducted for remote sensing image change detection landearth view image retrieval build-ing change detection dataset(LEVIR-CD)and Sun Yat-sen University change detection dataset(SYSU-CD).Each dataset is split into 7∶1∶2 for training,validation,and testing,respectively,and each image is segmented into 256 × 256 pixels.This approach facilitated the processing and increased the generalization capability of the model.The binary cross-entropy(BCELoss)is used as the loss function,and the performance of the proposed method is evaluated using three metrics:Fl score(Fl),intersection over union(IoU),and overall accuracy(OA).The proposed method achieved FI scores of 91.51％and 82.19％,IoU of 84.35％and 69.76％,and OA of 99.14％and 91.99％on the LEVIR-CD and SYSU-CD datasets,respectively.Compared with previous classical methods,the model obtained the best change detection results,especially preserving additional details on the change boundary.Ablation experiments are performed on the same dataset to illustrate the effect of low-level skip-connections and channel and spatial exchange modules.The results showed that the channel and spatial exchange module substantially optimized the utilization and representation of spatiotemporal features in the network,while the low-level skip-connection compensated for the loss of detailed features in the downsampling process and further enhanced the feature learning capability of the network.Conclusion The network used channel and spatial exchange modules to increase the utilization and understanding of spatiotemporal features and low-level skip-connections to focus the model on local detailed features.Finally,a binary cross-entropy loss function at the output layer is utilized to achieve optimal change detection performance.Experiments show that the method proposed in this paper can improve the capability of recognizing changing regions while ensuring a light network,enhancing the detection performance of change detection in various environments and terrains.

In Magnetotellurics (MT), limited by the computing efficiency of traditional methods, time-consuming is an urgent research topic for complex models. In the past few years, artificial intelligence has been applied to speed up calculations of partial differential equations. Such artificial intelligence methods are also applied in MT. These MT inversion algorithms are based on designing forward models using artificial intelligence. They are used to replace the traditional forward modeling algorithms for speedup in the inverse algorithms. However, it didn't directly solve the inverse problem. A novel approach with the combination of the invertible neural network and Fourier neural operator is proposed in this paper. Due to the reversibility of the proposed approach, the inverse process is completed by training the forward process in the meantime. In addition, it can perform bidirectional training for higher accuracy of the inverse problem. The experimental results show that the approach can perform both forward simulation and inverse problem prediction well simultaneously.

Abstract Hypertension remains a leading cause of cardiovascular and kidney diseases. Failure to control blood pressure with ≥ 3 medications or control requiring ≥ 4 medications is classified as resistant hypertension (rHTN) and new therapies are needed to reduce the resulting increased risk of morbidity and mortality. Here, we report genetic evidence that relaxin family peptide receptor 2 (RXFP2) is associated with rHTN in men, but not in women. This study shows that adrenal gland gene expression of RXFP2 is increased in men with hypertension and the RXFP2 natural ligand, INSL3, increases adrenal steroidogenesis and corticosteroid secretion in human adrenal cells. To address the hypothesis that RXFP2 activation is an important mechanism in rHTN, we discovered and characterized small molecule and monoclonal antibody (mAb) blockers of RXFP2. The novel chemical entities and mAbs show potent, selective inhibition of RXFP2 and reduce aldosterone and cortisol synthesis and release. The RXFP2 mAbs have suitable rat pharmacokinetic profiles to evaluate the role of RXFP2 in the development and maintenance of rHTN. Overall, we identified RXFP2 activity as a potential new mechanism in rHTN and discovered RXFP2 antagonists for the future interrogation of RXFP2 in cardiovascular and renal diseases.

Low-Dose computer tomography (LDCT) is an ideal alternative to reduce radiation risk in clinical applications. Although supervised-deep-learning-based reconstruction methods have demonstrated superior performance compared to conventional model-driven reconstruction algorithms, they require collecting massive pairs of low-dose and norm-dose CT images for neural network training, which limits their practical application in LDCT imaging. In this paper, we propose an unsupervised and training data-free learning reconstruction method for LDCT imaging that avoids the requirement for training data. The proposed method is a post-processing technique that aims to enhance the initial low-quality reconstruction results, and it reconstructs the high-quality images by neural work training that minimizes the ℓ1-norm distance between the CT measurements and their corresponding simulated sinogram data, as well as the total variation (TV) value of the reconstructed image. Moreover, the proposed method does not require to set the weights for both the data fidelity term and the plenty term. Experimental results on the AAPM challenge data and LoDoPab-CT data demonstrate that the proposed method is able to effectively suppress the noise and preserve the tiny structures. Also, these results demonstrate the rapid convergence and low computational cost of the proposed method. The source code is available at https://github.com/linfengyu77/IRLDCT.

Deep learning techniques have been used to build velocity models (VMs) for seismic traveltime tomography and have shown encouraging performance in recent years. However, they need to generate labeled samples (i.e., pairs of input and label) to train the deep neural network (NN) with end-to-end learning, and the real labels for field data inversion are usually missing or very expensive. Some traditional tomographic methods can be implemented quickly, but their effectiveness is often limited by prior assumptions. To avoid generating and/or collecting labeled samples, we propose a novel method by integrating deep learning and dictionary learning to enhance the VMs with low resolution by using the traditional tomography-least square method (LSQR). We first design a type of shallow and simple NN to reduce computational cost followed by proposing a two-step strategy to enhance the VMs with low resolution: 1) warming up: an initial dictionary is trained from the estimation by LSQR through the dictionary learning method; 2) dictionary optimization: the initial dictionary obtained in the warming-up step will be optimized by the NN, and then it will be used to reconstruct high-resolution VMs with the reference slowness and the estimation by LSQR. Furthermore, we design a loss function to minimize traveltime misfit to ensure that NN training is label-free, and the optimized dictionary can be obtained after each epoch of NN training. We demonstrate the effectiveness of the proposed method through the numerical tests on both synthetic and field data.

The nonlinear conjugate gradient (NLCG) algorithm is one of the popular linearized methods used to solve the frequency-domain electromagnetic (EM) geophysical inverse problem. During NLCG iterations, the model gradient guides the searching direction while the line-search algorithm determines the step length of each iteration. Normally, the line search requires solving the corresponding forward problem a few times. Since line search is usually computationally inefficient, we introduce the adaptive gradient descent (AGD) algorithm to accelerate solving the frequency-domain EM inverse problem within the linearized framework. The AGD algorithm is a variant of the classical gradient descent method and has been well-developed and widely used in deep learning. Rather than the time-consuming line search, its core idea is to algebraically manipulate the cumulative gradients and updates of the model from previous iterations to estimate the model parameter variables at the current iteration. For the inversion of magnetotelluric (MT) data, we here designed and implemented a framework using the AGD algorithm combined with the cool-down scheme to tune the regularization parameter. To improve the convergence performance of the AGD algorithm [specifying to Adam and root-mean-square propagation (RMSProp)], we proposed a tolerance strategy which has been tested numerically. To optimize the global learning rate, we carried out some comparative trials in the proposed inversion framework. The inverted results of synthetic and real-world data showed that both the AGD algorithms (Adam and RMSProp) can recover comparable results and save more than a third of CPU time compared with the NLCG algorithm.

Based on the actual demand of the public for 3D content generation technology and targeting the problems existing in traditional 2D to 3D image conversion methods, we propose a 2D to 3D pedestrian detection algorithm based on deep learning. The algorithm uses object segmentation to first extract the human body's 2D outline from the input image. It then combines shallow features of the image to use a fully connected layer to map latent vectors to low-dimensional mesh vertex feature vectors, ultimately generating a triangle mesh for the 3D model. By alternately using SpiralConv and upsampling operations, the 3D mesh of the pedestrian can be generated more accurately. Experimental results show that the algorithm can accurately construct a 3D model of a pedestrian and has real-time feedback capability. The algorithm is suitable for real-time monitoring scenarios such as action recognition. Compared to traditional 2D to 3D image conversion methods, the algorithm uses deep learning to automatically learn features, which has higher accuracy and transferability. Additionally, the algorithm extracts 2D contour information of the human body using object segmentation, which improves the accuracy of generating 3D meshes. We believe that this algorithm has an important role to play in the development of 3D content generation technology.

Deep learning techniques have been widely used for semantic change detection (SCD) of remote sensing images (RSIs) and have shown encouraging performance. In this letter, we propose a novel neural network by embedding the difference enhancement (DE) module into the adjacent layers of ResNet for SCD of RSIs (DESNet), which can pay more attention to the changes of bitemporal RSIs. Furthermore, we deploy the module of multiscale parallel sampling spatial–spectral nonlocal (SSN) after feature extraction, which can effectively improve the robustness to large-scale changes and the integrity of the changed objects by fusing global features that sampled from the multiscale feature space. The experimental tests demonstrate that our DESNet can achieve state-of-the-art accuracy on the SECOND dataset and the Landsat-SCD dataset.

This paper introduces an approximate nuclear norm based matrix regression projection(ANMRP)model,an adaptive graph embedding method,for feature extraction of hyperspectral images.The ANMRP utilizes an approximate NMR model to construct an adaptive neighborhood map between samples.The globally optimal weight matrix is obtained by optimizing the approximate NMR model using fast alternating direction method of multipliers(ADMM).The optimal projection matrix is then determined by maximizing the ratio of the local scatter matrix to the total scatter matrix,al-lowing for the extraction of discriminative features.Experimental results demonstrate the effectiveness of ANMRP compared to related methods.

For seismic random noise attenuation, deep learning has attracted much attention and achieved promising performance. However, compared with conventional methods, the denoising performance of supervised learning-based methods heavily depends on massive training samples with high-quality labeled data, which makes their generalization capabilities limited. Even though deep neural networks (DNNs) usually outperform the conventional denoising methods, their performance is not guaranteed since neural networks still lack good mathematical interpretability at present. To alleviate the dependency on labeled data and explore insights into the denoising system, we proposed an unsupervised denoising method based on model-based deep learning, which combined domain knowledge and a data-driven method. We designed a network based on the modified iterative soft threshold algorithm (ISTA), which omitted the soft threshold to alleviate uncertainties introduced by empirically selected thresholds. In this network, we set the dictionary and code as trainable parameters. A loss function with a smooth penalty was designed to ensure that the network training can be implemented in an unsupervised manner. In the proposed method, we set the denoised result by $f-x$ deconvolution as the input for our network, and the further denoised data can be obtained after each epoch of the training, which means that our method does not need the testing procedure. Experiments on synthetic and field seismic data demonstrate that our method exhibits competitive performance compared to the conventional, supervised, and unsupervised methods, including $f-x$ deconvolution, curvelet, the Denoising Convolutional Neural Network (DnCNN), and the integration of neural network and Block-matching and 3-D filtering method (NN + BM3D).

The efficiency of solving geophysical inverse problem largely relies on the efficiency of solving the corresponding forward problem. As for electromagnetic (EM) data forward modeling in frequency domain, the conventional numerical methods, e.g., finite difference method (FDM), discretize the governing equations resulting in a large linear system which is usually expensive to solve. Meanwhile, for inversion iteration, we normally do not need to solve the forward problem in high precision. Thus, a rapid surrogate modeling approach which uses the neural network is promising for replacing the forward modeling module in the inversion scheme. Here, we proposed an algorithm which uses the neural operator to solve the EM data modeling problem in the frequency domain. To develop a surrogate model for EM data forward problem, we introduce an extended Fourier neural operator (EFNO) that enables the calculation at least 100 times faster than the conventional FDM solver while maintaining good precision. Moreover, by adding a subnetwork the proposed neural operator has good generalization which has the capacity of predicting solution at any site locations and frequencies. Due to the discretization invariance of Fourier neural operator, the neural operator trained on coarse grids can easily transfer to fine grids with only retraining part of parameters, resulting in a super-resolution prediction capability. We test our proposed method with 2-D and 3-D magnetotelluric (MT) data modeling problems, demonstrating that the EFNO has great potentials for severing as a general rapid surrogate forward solver in EM data inversion scheme.

SUMMARYGeophysical inverse problems are commonly ill-posed and require proper regularization to gain acceptable solutions. Adopting constraints on the smoothness and/or specified structures of an inverted geophysical model requires the implementation of regularization terms when either structured or unstructured meshes are used. Spatial gradients of the model parameters along axial or arbitrary directions are thus required. However, the calculation of spatial gradients on unstructured meshes is not straightforward since the interfaces between adjacent model cells (i.e. common edges or common faces) are orientated individually and usually are not perpendicular to the axial directions. Depending on the algorithm used, an uneven performance of the regularization is observed on unstructured meshes. To enforce effective and robust regularization terms for geophysical inversions on the unstructured meshes, we propose algorithms for constructing the smoothness and structural similarity operators that take advantage of the inherent merits of the unstructured meshes. Following a detailed introduction of the general inversion formula that we adopted, the smoothness and reference model constraints on triangular and tetrahedral meshes are proposed based on the neighbouring relationships between different model cells within the meshes. Particularly, a quasi-cross-gradient formulation is derived for triangular meshes suitable for the joint inversion of different kinds of geophysical data. Compared to existing algorithms, the new smoothness operator presents an equal or better performance for constraining the model roughness. In addition, the operator exploits the preferred elongation directions of the underground structures by performing varied constraints in different directions. Furthermore, the other new operator could effectively measure structural information of the inverted model even if the algorithms have incorporated sophisticated constraints from other geophysical or geological data. Demonstrated with the applications on synthetic examples, the new algorithms provide advanced regularization techniques for conducting geophysical inversions using unstructured meshes.

ABSTRACT Impedance is a physical parameter that plays an important role in seismic data processing and interpretation. A relative impedance perturbation (the ratio of the impedance perturbation and the impedance for the background models) imaging method in depth domain based on the reflection wave equation is proposed. Under the small perturbation assumption, primary wave and high‐frequency approximation condition, a linear propagation equation of the primary reflection waves based on the relative impedance perturbation was first derived. On this basis, we further derived the imaging formula of the relative impedance perturbation using a linear inversion theory. Then, the source–receiver bidirectional illumination compensation was used to improve the image quality of the subsurface structures. The image result obtained by this method can be used to estimate the relative impedance perturbation. In the angle domain, the extracted near‐angle‐domain image gather with amplitude compensation can estimate the relative impedance perturbation, and the far‐angle image gather provides the estimation of the relative velocity perturbation (the ratio of the velocity perturbation and the background velocity). Finally, several numerical tests demonstrate the effectiveness of the method.

Impedance is a physical parameter that plays an important role in seismic data processing and interpretation. A relative impedance perturbation (the ratio of the impedance perturbation and the impedance for the background models) imaging method in depth domain based on the reflection wave equation is proposed. Under the small perturbation assumption, primary wave and high-frequency approximation condition, a linear propagation equation of the primary reflection waves based on the relative impedance perturbation was first derived. On this basis, we further derived the imaging formula of the relative impedance perturbation using a linear inversion theory. Then, the source–receiver bidirectional illumination compensation was used to improve the image quality of the subsurface structures. The image result obtained by this method can be used to estimate the relative impedance perturbation. In the angle domain, the extracted near-angle-domain image gather with amplitude compensation can estimate the relative impedance perturbation, and the far-angle image gather provides the estimation of the relative velocity perturbation (the ratio of the velocity perturbation and the background velocity). Finally, several numerical tests demonstrate the effectiveness of the method.

Deep learning has achieved a great success in various fields in recent years. As a powerful tool, deep learning is used to interpolate the gravity and magnetic data with block lack of data in this paper. We firstly designed a deep convolution neural network (CNN), and mask dataset was created to simulate the block lack in the data. The network was trained by using the earth’s gravity anomalies dataset (bouguer, isostatic and surface free-air) and the earth magnetic anomaly dataset. We demonstrated the feasibility of this approach by numerical experiments, which indicated that this method is able to effectively recover the gravity and magnetic data with block lack. The experiments also demonstrated that CNN gains competitive performance compared with traditional interpolation methods. Presentation Date: Wednesday, September 18, 2019 Session Start Time: 9:20 AM Presentation Time: 10:35 AM Location: Poster Station 1 Presentation Type: Poster