Search

Arbitrary style transfer (AST) has garnered considerable attention for its ability to transfer styles infinitely. Although existing methods have achieved impressive results, they may overlook style consistencies and fail to capture crucial style patterns, leading to inconsistent style transfer (ST) caused by minor disturbances. To tackle this issue, we conduct a mathematical analysis of inconsistent ST and develop a style inconsistency measure (SIM) to quantify the inconsistencies between generated images. Moreover, we propose a consistent AST (CAST) framework that effectively captures and transfers essential style features into content images. The proposed CAST framework incorporates an intersection-of-union-preserving crop (IoUPC) module to obtain style pairs with minor disturbance, a self-attention (SA) module to learn the crucial style features, and a style inconsistency loss regularization (SILR) to facilitate consistent feature learning for consistent stylization. Our proposed framework not only provides an optimal solution for consistent ST but also outperforms existing methods when embedded into the CAST framework. Extensive experiments demonstrate that the proposed CAST framework can effectively transfer style patterns while preserving consistency and achieve the state-of-the-art performance.

Multiview clustering (MVC), which can dexterously uncover the underlying intrinsic clustering structures of the data, has been particularly attractive in recent years. However, previous methods are designed for either complete or incomplete multiview only, without a unified framework that handles both tasks simultaneously. To address this issue, we propose a unified framework to efficiently tackle both tasks in approximately linear complexity, which integrates tensor learning to explore the inter-view low-rankness and dynamic anchor learning to explore the intra-view low-rankness for scalable clustering (TDASC). Specifically, TDASC efficiently learns smaller view-specific graphs by anchor learning, which not only explores the diversity embedded in multiview data, but also yields approximately linear complexity. Meanwhile, unlike most current approaches that only focus on pair-wise relationships, the proposed TDASC incorporates multiple graphs into an inter-view low-rank tensor, which elegantly models the high-order correlations across views and further guides the anchor learning. Extensive experiments on both complete and incomplete multiview datasets clearly demonstrate the effectiveness and efficiency of TDASC compared with several state-of-the-art techniques.

Due to the high cost of pixel-level labels required for fully-supervised semantic segmentation, weakly-supervised segmentation has emerged as a more viable option recently. Existing weakly-supervised methods tried to generate pseudo-labels without pixel-level labels for semantic segmentation, but a common problem is that the generated pseudo-labels contain insufficient semantic information, resulting in poor accuracy. To address this challenge, a novel method is proposed, which generates class activation/attention maps (CAMs) containing sufficient semantic information as pseudo-labels for the semantic segmentation training without pixel-level labels. In this method, the attention-transfer module is designed to preserve salient regions on CAMs while avoiding the suppression of inconspicuous regions of the targets, which results in the generation of pseudo-labels with sufficient semantic information. A pixel relevance focused-unfocused module has also been developed for better integrating contextual information, with both attention mechanisms employed to extract focused relevant pixels and multi-scale atrous convolution employed to expand receptive field for establishing distant pixel connections. The proposed method has been experimentally demonstrated to achieve competitive performance in weakly-supervised segmentation, and even outperforms many saliency-joined methods.

Subspace learning has been widely applied for joint feature extraction and dimensionality reduction, demonstrating significant efficacy. Numerous subspace learning methods with diverse assumptions regarding the criteria for the target subspaces have been developed to obtain compact and interpretable data representations. However, when applied to image data, existing methods fail to fully exploit the inherent correlations within the image set. This paper proposes a Robust Discriminative t-Linear Subspace Learning model (RDtSL) to tackle this issue using t-product. The model mainly has four strengths: 1) Taking advantage of t-product, RDtSL learns the projection basis directly from the image set while fully exploiting its internal correlations; 2) Based on its energy preservation module, RDtSL retains the primary energy of samples in the learned subspace, maintaining satisfactory performance even with low subspace dimensions; 3) Class-distinctive features are effectively preserved in the learned representations due to the incorporation of the classification module; 4) Relying on its graph embedding module, RDtSL learns an affinity graph of samples adaptively to enrich the data representations with locality and similarity information. The harmonious balance maintained between the three proposed modules helps RDtSL learn discriminative and informative data representations. We also develop an iterative algorithm to solve RDtSL. Extensive experiments on benchmark databases demonstrate the superiority of the proposed model.

For autonomous ground vehicles, global localization with 3D LiDAR is an indispensable part of tasks such as navigation. Usually, global localization using LiDAR is subdivided into two sub-problems, place recognition and global registration. For place recognition, the recent emerging schemes based on deep learning either rely on 3D convolution with high complexity or need to learn features from various forward perspectives. To mitigate this, we propose a model with roll-pitch-yaw invariance that represents point clouds as probabilistic voxels and generates occupancy grids from a bird’s-eye view, fulfilling robust place recognition by learning aggregated embeddings from a fixed perspective. For low-overlap global registration, the traditional handcraft feature-based methods are mostly limited to dense object-level point clouds, while the state-of-the-art learning-based approaches often rely on complex 3D convolution and additional feature association learning. To fill this gap to some extent, we propose to estimate the relative roll-pitch angles and vertical translation by fitting and aligning the ground plane of the point clouds and to determine the horizontal translations and yaw angle by matching their projected occupancy grids. Extensive experiments corroborate the superior recall and generalization ability of our place recognition model, as well as the advanced success rate and accuracy of our 3D registration approach. Especially in the recognition and registration of hard samples, our results far exceed those of our counterparts by large margins. To ensure full reproducibility, the relevant codes and data are made available online at https://cslinzhang.github.io/GLoc/GLoc.html.

Recently, it has been shown that adversaries can reconstruct images from SIFT features through reverse attacks. However, the images reconstructed by existing reverse attack methods suffer from information loss and are unable to sufficiently reveal the private contents of the original images. In this paper, a two-stage deep reverse attack model called Coarse-to-Fine Generative Adversarial Network (CFGAN) is proposed to more deeply explore the information in SIFT features and further demonstrate the risk of privacy leakage associated with SIFT features. Specifically, the proposed model consists of two sub-networks, namely coarse net and fine net. The coarse net is developed to restore coarse images using SIFT features, while the fine net is responsible for refining the coarse images to obtain better reconstruction results. To effectively leverage the information contained in SIFT features, an efficient fusion strategy based on the AdaIN operation is designed in the fine net. Additionally, we introduce a new loss function called sift loss that enhances the color fidelity of reconstructed images. Extensive experiments conducted on various datasets verify that the proposed CFGAN performs favorably against state-of-the-art methods. The reconstructed images exhibit better visual quality, less texture distortion, and higher color fidelity. Source code is available at https://github.com/HITLiXincodes/CFGAN.

The harmonic neural network (HNN) learns a combination of discrete cosine transform (DCT) filters to obtain an integrated feature from all spectra in the frequency domain. HNN, however, faces two challenges in learning and inference processes. First, the spectrum feature learned by HNN is insufficient and limited because the number of DCT filters is much smaller than that of feature maps. In addition, the number of parameters and the computation costs of HNN are significantly high because the intermediate spectrum layers are expanded multiple times. These two challenges will severely harm the performance and efficiency of HNN. To solve these problems, we first propose the compound DCT (C-DCT) filters integrating the nearest DCT filters to retrieve rich spectrum features to improve the performance. To significantly reduce the model size and computation complexity for improving the efficiency, the shared reconstruction filter is then proposed to share and dynamically drop the meta-filters in every frequency branch. Integrating the C-DCT filters with the shared reconstruction filters, the efficient harmonic network (EH-Net) is introduced. Extensive experiments on different datasets demonstrate that the proposed EH-Nets can effectively reduce the model size and computation complexity while maintaining the model performance. The code has been released at https://github.com/zhangle408/EH-Nets.

With the rapid progress of deepfake techniques in recent years, facial video forgery can generate highly deceptive video content and bring severe security threats. And detection of such forgery videos is much more urgent and challenging. Most existing detection methods treat the problem as a vanilla binary classification problem. In this article, the problem is treated as a special fine-grained classification problem since the differences between fake and real faces are very subtle. It is observed that most existing face forgery methods left some common artifacts in the spatial domain and time domain, including generative defects in the spatial domain and interframe inconsistencies in the time domain. And a spatial-temporal model is proposed which has two components for capturing spatial and temporal forgery traces from a global perspective, respectively. The two components are designed using a novel long-distance attention mechanism. One component of the spatial domain is used to capture artifacts in a single frame, and the other component of the time domain is used to capture artifacts in consecutive frames. They generate attention maps in the form of patches. The attention method has a broader vision which contributes to better assembling global information and extracting local statistic information. Finally, the attention maps are used to guide the network to focus on pivotal parts of the face, just like other fine-grained classification methods. The experimental results on different public datasets demonstrate that the proposed method achieves state-of-the-art performance, and the proposed long-distance attention method can effectively capture pivotal parts for face forgery.

Variable selection methods aim to select the key covariates related to the response variable for learning problems with high-dimensional data. Typical methods of variable selection are formulated in terms of sparse mean regression with a parametric hypothesis class, such as linear functions or additive functions. Despite rapid progress, the existing methods depend heavily on the chosen parametric function class and are incapable of handling variable selection for problems where the data noise is heavy-tailed or skewed. To circumvent these drawbacks, we propose sparse gradient learning with the mode-induced loss (SGLML) for robust model-free (MF) variable selection. The theoretical analysis is established for SGLML on the upper bound of excess risk and the consistency of variable selection, which guarantees its ability for gradient estimation from the lens of gradient risk and informative variable identification under mild conditions. Experimental analysis on the simulated and real data demonstrates the competitive performance of our method over the previous gradient learning (GL) methods.

Optimizing the computational efficiency of the artificial neural networks is crucial for resource-constrained platforms like autonomous driving systems. To address this challenge, we proposed a Lightweight Context-aware Network (LCNet) that accelerates semantic segmentation while maintaining a favorable trade-off between inference speed and segmentation accuracy in this paper. The proposed LCNet introduces a partial-channel transformation (PCT) strategy to minimize computing latency and hardware requirements of the basic unit. Within the PCT block, a three-branch context aggregation (TCA) module expands the feature receptive fields, capturing multiscale contextual information. Additionally, a dual-attention-guided decoder (DD) recovers spatial details and enhances pixel prediction accuracy. Extensive experiments on three benchmarks demonstrate the effectiveness and efficiency of the proposed LCNet model. Remarkably, a smaller model LCNet $_{3\_{}7}$ achieves 73.8% mIoU with only 0.51 million parameters, with an impressive inference speed of $\sim $ 142.5 fps and $\sim $ 9 fps using a single RTX 3090 GPU and Jetson Xavier NX, respectively, on the Cityscapes test set at $1024\times 1024$ resolution. A more accurate version of the LCNet $_{3\_{}11}$ can achieve 75.8% mIoU with 0.74 million parameters at $\sim $ 117 fps inference speed on Cityscapes at the same resolution. Much faster inference speed can be achieved at smaller image resolutions. LCNet strikes a great balance between computational efficiency and prediction capability for mobile application scenarios. The code is available at https://github.com/lztjy/LCNet.

Owing to the inherent complementarity among LiDAR, camera, and IMU, a growing effort has been paid to laser-visual-inertial SLAM recently. The existing approaches, however, are limited in two aspects. First, at the front-end, they usually employ a discrete-time representation that requires high-precision hardware/software synchronization and are based on geometric laser features, leading to low robustness and scalability. Second, at the backend, visual loop constraints suffer from scale ambiguity and the sparseness of the point cloud deteriorates the scan-to-scan loop detection. To solve these problems, for the front-end, we propose a continuous-time laser-visual-inertial odometry which formulates the carrier trajectory in continuous time, organizes point clouds in probabilistic submaps, and jointly optimizes the loss terms of laser anchors, visual reprojections, and IMU readings, achieving accurate pose estimation even with fast motion or in unstructured scenes where it is difficult to extract meaningful geometric features. At the backend, we propose building 5-DoF laser constraints by matching projected 2D submaps and 6-DoF visual constraints via laser-aided visual relocalization, ensuring mapping consistency in large-scale scenes. Results show that our framework achieves high-precision estimation and is more robust than its counterparts when the carrier works in large scenes or with fast motion. The relevant codes and data are open-sourced at https://cslinzhang.github.io/Ct-LVI/Ct-LVI.html.

In this article, the problem of impulse noise image restoration is investigated. A typical way to eliminate impulse noise is to use an $L_{1}$ norm data fitting term and a total variation (TV) regularization. However, a convex optimization method designed in this way always yields staircase artifacts. In addition, the $L_{1}$ norm fitting term tends to penalize corrupted and noise-free data equally, and is not robust to impulse noise. In order to seek a solution of high recovery quality, we propose a new variational model that integrates the nonconvex data fitting term and the nonconvex TV regularization. The usage of the nonconvex TV regularizer helps to eliminate the staircase artifacts. Moreover, the nonconvex fidelity term can detect impulse noise effectively in the way that it is enforced when the observed data is slightly corrupted, while is less enforced for the severely corrupted pixels. A novel difference of convex functions algorithm is also developed to solve the variational model. Using the variational method, we prove that the sequence generated by the proposed algorithm converges to a stationary point of the nonconvex objective function. Experimental results show that our proposed algorithm is efficient and compares favorably with state-of-the-art methods.

Numerous state-of-the-art transformer-based techniques with self-attention mechanisms have recently been demonstrated to be quite effective in the classification of hyperspectral images (HSIs). However, traditional transformer-based methods severely suffer from the following problems when processing HSIs with three dimensions: (1) processing the HSIs using 1D sequences misses the 3D structure information; (2) too expensive numerous parameters for hyperspectral image classification tasks; (3) only capturing spatial information while lacking the spectral information. To solve these problems, we propose a novel Quaternion Transformer Network (QTN) for recovering self-adaptive and long-range correlations in HSIs. Specially, we first develop a band adaptive selection module (BASM) for producing Quaternion data from HSIs. And then, we propose a new and novel quaternion self-attention (QSA) mechanism to capture the local and global representations. Finally, we propose a new and novel transformer method, i.e., QTN by stacking a series of QSA for hyperspectral classification. The proposed QTN could exploit computation using Quaternion algebra in hypercomplex spaces. Extensive experiments on three public datasets demonstrate that the QTN outperforms the state-of-the-art vision transformers and convolution neural networks.

Aiming at the large distortion and low tampering localization accuracy of the existing semi-fragile reversible watermarking for 3D mesh models, a novel semi-fragile reversible watermarking for 3D models using spherical crown volume division is proposed. The crown volume of a sphere is divided to reduce the embedding distortion. The possible geometric and topological transformations are separately considered in the watermark generation, and the vertices of the one-ring neighbourhood are grouped to improve the tampering localization accuracy. Experimental results show that the proposed scheme can achieve better localization accuracy and lower embedding distortion than some state-of-the-art algorithms. It has good potential for the applications in integrity authentication for 3D mesh models.

Image steganographic methods based on encoder-decoder model with end-to-end network architecture recently have been proposed. However, in steganographic applications, the feature map (called stego matrix) generated by the encoder needs to be rounded as a real stego image for the receiver. The loss of precision by rounding stego matrix leads to the decline in the accuracy of extracted secret messages. The challenge of using end-to-end network to preserve robustness against rounding operation is that it is non-differentiable. In this paper, we propose an anti-rounding image steganography method with separable fine-tuning network architecture which includes the joint training stage (JT-stage) and the separable fine-tuning stage (SF-stage). Firstly, in JT-stage, an embedded generator and a stego matrix extractor are jointly learned without rounding operation. Utilizing concatenation in embedded generator can realistically fuse cover image and secret messages. And the multi-scale fusion block and residual dense block in stego matrix extractor can make secret messages more correctly decoded. Moreover, the discriminator is constructed by generative adversarial nets (GAN) in JT-stage to effectively improve the authenticity and steganalysis security. Then, in SF-stage, the embedded generator is frozen, and the stego matrix is obtained and rounded as a stego image. A stego image extractor is constructed by fine-tuning the layers of the stego matrix extractor to improve the accuracy of message extraction. As the loss will not backpropagate in the embedded generator, the non-differentiability of rounding operation can be offset. Experiments show that the proposed separation fine-tuning network is robust to rounding operation, and effectively reduces the degradation of the image quality and steganalysis performance.

So far, researchers have proposed many forensics tools to protect the authenticity and integrity of digital information. However, with the explosive development of machine learning, existing forensics tools may compromise against new attacks anytime. Hence, it is always necessary to investigate anti-forensics to expose the vulnerabilities of forensics tools. It is beneficial for forensics researchers to develop new tools as countermeasures. To date, one of the potential threats is the generative adversarial networks (GANs), which could be employed for fabricating or forging falsified data to attack forensics detectors. In this article, we investigate the anti-forensics performance of GANs by proposing a novel model, the ExS-GAN, which features an extra supervision system. After training, the proposed model could launch anti-forensics attacks on various manipulated images. Evaluated by experiments, the proposed method could achieve high anti-forensics performance while preserving satisfying image quality. We also justify the proposed extra supervision via an ablation study.

High-dimensional (HD) chaotic map has wide applications in various research fields such as neural networks and secure communication. Designing HD chaotic maps with expected dynamics and robust hyperchaotic behaviors is an interesting but challenging topic. In this article, we propose an $n$ -dimensional hyperchaotic map $(n\text{D}$ -HCM) generation method on the basis of the Gershgorin-type theorem. First, the general form of the proposed $n\text{D}$ -HCM is built using $n$ parametric polynomials. Then, the entity and coefficient parameter matrices are configured according to the Gershorin-type theorem. Theoretical analysis shows that the generated $n\text{D}$ -HCM has $n$ positive Lyapunov exponents and thus can show robust hyperchaotic behaviors. Two examples of hyperchaotic map with specified equations are provided and their properties are analyzed to show the availability of the proposed method. Performance evaluations display that our $n\text{D}$ -HCM possesses abundant properties and complex behaviors, and it can outperform some representative HD chaotic maps. Moreover, to show the application of our $n\text{D}$ -HCM, we apply it to a secure communication scheme and the experimental results exhibit that it shows much better performance than these representative HD chaotic maps in resisting transmission noise.

Motivated by the advance of deep learning methods, deep unfolding methods such as deep convolutional dictionary learning have achieved great success in image denoising tasks. The main advantages are inheriting both the merits of deep learning (strong learning capacity) and traditional machine learning (powerful interpretable capacity). We observe that the update of dictionaries and coefficients is highly correlated with the previous iterative stage information for deep unfolding-based methods. However, most existing deep convolutional dictionary learning methods deal with each iteration step individually, ignoring the inner-memory within the stage and cross-memory across the stages. To alleviate these issues, we propose a dynamic inner-cross memory augmented attentional dictionary learning (M2ADL) network with attention guided residual connection module, which utilizes the previous important stage features such that better uncovering the inner-cross information. Specifically, the proposed inner-cross memory fully utilizes the previous stage’s hidden and last-layer information to learn the dictionary. In addition, we develop a dual attention-guided residual connection module to well exploit the deep feature learning ability to capture the spatial-spectral attention across the deep tensor-based features. Considerable experiments on both synthetic and real image datasets demonstrate the superiority of the proposed method over other state-of-the-art methods.

Monitoring and forecasting of sintering temperature (ST) is vital for safe, stable, and efficient operation of rotary kiln production process. Due to the complex coupling and time-varying characteristics of process data collected by the distributed control system, its long-range prediction remains a challenge. In this article, we propose a multivariate time series forecasting model based on dynamic spatio-temporal graph attention network (GAT) to model time-varying spatio-temporal correlation between the process data and perform long-range forecasting of ST. Aiming at the problem that there is no preset graph structure for multivariate data, we first propose an adaptive adjacency matrix generation algorithm to construct an elementary graph structure for the process data. Then, we design a spatio-temporal graph attention module, which consists of a multihead GAT for extracting time-varying spatial features and a gated dilated convolutional network for temporal features. Finally, considering the different time delay and rhythm of each process variable, we use dynamic system analysis to estimate the delay time and rhythm of each variable to guide the selection of dilation rates in dilated convolutional layers. The application results based on actual data show that the method has high prediction accuracy, and has broad application prospects in industrial processes.

Decoupled Few-shot learning (FSL) is an effective methodology that deals with the problem of data-scarce. Its standard paradigm includes two phases: (1) Pre-train. Generating a CNN-based feature extraction model (FEM) via base data. (2) Meta-test. Employing the frozen FEM to obtain the novel data features, then classifying them. Obviously, one crucial factor, the category gap, prevents the development of FSL, i.e., it is challenging for the pre-trained FEM to adapt to the novel class flawlessly. Inspired by a common-sense theory: the FEMs based on different strategies focus on different priorities, we attempt to address this problem from the multi-view feature collaboration (MVFC) perspective. Specifically, we first denoise the multi-view features by subspace learning method, then design three attention blocks (loss-attention block, self-attention block and graph-attention block) to balance the representation between different views. The proposed method is evaluated on four benchmark datasets and achieves significant improvements of 0.9%-5.6% compared with SOTAs.

Copy-move forgery can be used for hiding certain objects or duplicating meaningful objects in images. Although copy-move forgery detection has been studied extensively in recent years, it is still a challenging task to distinguish between the source and the target regions in copy-move forgery images. In this paper, a convolutional neural network-transformer based generative adversarial network (CNN-T GAN) is proposed to distinguish the source and target regions in a copy-move forged image. A generator is first utilized to generate a mask that is similar to the groundtruth mask. Then, a discriminator is trained to discriminate the true image pairs from the false ones. When the discriminator cannot discriminate the true/false image pairs accurately, the generator can be used to obtain the final localization maps of copy-move forgery. In the generator, convolutional neural network (CNN) and transformer are exploited to extract the local features and global representations in copy-move forgery images, respectively. In addition, feature coupling layers are designed to integrate the features in CNN branch and transformer branch in an interactive way. Finally, a new Pearson correlation layer is introduced to match the similarity features in source and target regions, which can improve the performance of copy-move forgery localization, especially the localization performance on source regions. To the best of our knowledge, this is the first work to utilize transformer for feature extraction in copy-move forgery localization. The proposed method can not only detect the copy-move regions, but also distinguish the source and target regions. Extensive experimental results on several commonly used copy-move datasets have shown that the proposed method outperforms the state-of-the-art methods for copy-move detection.

Due to the physical nature of color images, color image processing such as denoising and inpainting has shown extensive and versatile possibilities over grayscale image processing. The monochromatic and the concatenation model have been widely used to process color images by processing each color channel independently or concatenating three color channels as one unified one and then used existing grayscale image processing methods directly without specific operations. These above schemes, however, have some limitations: (1) they would destroy the inherent correlation among three color channels since they cannot represent color images holistically; (2) they usually focus on one specific handcrafted prior such as smoothness, low-rankness, or even deep prior and thus failing to fuse deep and handcrafted priors of color images flexibly. To conquer these limitations, we propose one unified model to integrate deep prior and low-rank quaternion prior (DLRQP) for color image processing under the plug-and-play (PnP) framework. Specifically, the quaternion representation with low-rank constraint is introduced to denote the color image in a holistic way and one advanced denoiser is adopted to explore the deep prior in an iterative process. To tightly approximate the quaternion rank, one nonconvex penalty function is further utilized. We derive an alternate iterative approach to tackle the proposed model. We empirically demonstrate that our model can achieve superior performance over existing methods on both color image denoising and inpainting tasks.

The remote sensing (RS) image super-resolution (SR) algorithm aims to reconstruct a high-resolution (HR) image with rich texture details from a given low-resolution (LR) image, improving the spatial resolution. It has been widely concerned in RS image processing and application. Most current deep-learning-based methods rely on paired training datasets. However, most datasets are often based on bicubic degradation. This single construction way limits the performance of the pretrained network. Moreover, SR is an ill-posed problem in that multiple SR images are constructed from a single LR input. This article proposes a region-aware network (RAN) for RS image SR to alleviate the above issues. First, we introduce the contrastive learning strategy to mine the latent degraded representation of the image and serve as the prior knowledge of the network. Considering the RS images are acquired in specific scenes that have apparent self-similarity. Then, we propose a region-aware module (RAM) based on attention mechanisms and the graph neural network to explore region information and cross-patch self-similarity. Extensive experiments have demonstrated that the proposed RAN adapts to RS image SR tasks with various degradations and performs better in constructing texture information.

Real-time semantic segmentation is widely used in autonomous driving and robotics. Most previous networks achieved great accuracy based on a complicated model involving mass computing. The existing lightweight networks generally reduce the parameter sizes by sacrificing the segmentation accuracy. It is critical to balance the parameters and accuracy for real-time semantic segmentation. In this article, we propose a lightweight multiscale-feature-fusion network (LMFFNet) mainly composed of three types of components: split-extract-merge bottleneck (SEM-B) block, feature fusion module (FFM), and multiscale attention decoder (MAD), where the SEM-B block extracts sufficient features with fewer parameters. FFMs fuse multiscale semantic features to effectively improve the segmentation accuracy and the MAD well recovers the details of the input images through the attention mechanism. Without pretraining, LMFFNet-3-8 achieves 75.1% mean intersection over union (mIoU) with 1.4 M parameters at 118.9 frames/s using RTX 3090 GPU. More experiments are investigated extensively on various resolutions on other three datasets of CamVid, KITTI, and WildDash2. The experiments verify that the proposed LMFFNet model makes a decent tradeoff between segmentation accuracy and inference speed for real-time tasks. The source code is publicly available at https://github.com/Greak-1124/LMFFNet.

In this letter, we present a single mixed-integer linear programming (MILP) model for high-stealth false data attacks (FDAs) on overloading a set of lines by injecting stealthy false data. The proposed model reveals that an intelligent attacker is able to deliberately construct a valid attack vector to overload multiple transmission lines while hiding it among normal data to evade advanced anomaly detection methods. In addition, the proposed cyber-attack mode can help the attacker optimally select the targeted lines. Simulation results on multiple large-scale test systems validate the effectiveness of the proposed approach.

Recently, prototypical network-based few-shot learning (FSL) has been introduced for small-sample hyperspectral image (HSI) classification and has shown good performance. However, existing prototypical-based FSL methods have two problems: prototype instability and domain shift between training and testing datasets. To solve these problems, we propose a refined prototypical contrastive learning network for FSL (RPCL-FSL) in this article, which incorporates supervised contrastive learning (CL) and FSL into an end-to-end network to perform small-sample HSI classification. To stabilize and refine the prototypes, RPCL-FSL imposes triple constraints on prototypes of the support set, i.e., CL-, self-calibration (SC)-, and cross-calibration (CC)-based constraints. The CL module imposes an internal constraint on the prototypes aiming to directly improve the prototypes using support set samples in the CL framework, and the SC and CC modules impose external constraints on the prototypes by using the prediction loss of support set samples and the query set prototypes, respectively. To alleviate a domain shift in the FSL, a fusion training strategy is designed to reduce the feature differences between training and testing datasets. Experimental results on three HSI datasets demonstrate that the proposed RPCL-FSL outperforms existing state-of-the-art deep learning and FSL methods.

Few-shot learning (FSL), purposing to resolve the problem of data-scarce, has attracted considerable attention in recent years. A popular FSL framework contains two phases: (i) the pre-train phase employs the base data to train a CNN-based feature extractor. (ii) the meta-test phase applies the frozen feature extractor to novel data (novel data has different categories from base data) and designs a classifier for recognition. To correct few-shot data distribution, researchers propose Semi-Supervised Few-Shot Learning (SSFSL) by introducing unlabeled data. Although SSFSL has been proved to achieve outstanding performances in the FSL community, there still exists a fundamental problem: the pre-trained feature extractor cannot adapt to the novel data flawlessly due to the cross-category setting. Usually, large amounts of noises are introduced to the novel feature. We dub it as Feature-Extractor-Maladaptive (FEM) problem. To tackle FEM, we make two efforts in this paper. First, we propose a novel label prediction method, Isolated Graph Learning (IGL). IGL introduces the Laplacian operator to encode the raw data to graph space, which helps reduce the dependence on features when classifying, and then project graph representation to label space for prediction. The key point is that: IGL can weaken the negative influence of noise from the feature representation perspective, and is also flexible to independently complete training and testing procedures, which is suitable for SSFSL. Second, we propose Graph Co-Training (GCT) to tackle this challenge from a multi-modal fusion perspective by extending the proposed IGL to the co-training framework. GCT is a semi-supervised method that exploits the unlabeled samples with two modal features to crossly strengthen the IGL classifier. We estimate our method on five benchmark few-shot learning datasets and achieve outstanding performances compared with other state-of-the-art methods. It demonstrates the effectiveness of our GCT.

Autoencoder (AE) is an unsupervised neural network framework for efficient and effective feature extraction. Most AE-based methods do not consider spatial information and band correlations for hyperspectral image (HSI) analysis. In addition, graph-based AE methods often learn discriminative representations with the assumption that connected samples share the same label and they cannot directly embed the geometric structure into feature extraction. To address these issues, in this paper, we propose a dual graph autoencoder (DGAE) to learn discriminative representations for HSIs. Utilizing the relationships of pair-wise pixels within homogenous regions and pair-wise spectral bands, DGAE first constructs the superpixel-based similarity graph with spatial information and band-based similarity graph to characterize the geometric structures of HSIs. With the developed dual graph convolution, more discriminative feature representations are learnt from the hidden layer via the encoder-decoder structure of DGAE. The main advantage of DGAE is that it fully exploits both the geometric structures of pixels with spatial information and spectral bands to promote nonlinear feature extraction of HSIs. Experiments on HSI datasets show the superiority of the proposed DGAE over the state-of-the-art methods. The source code of DGAE is available at https://github.com/ZhangYongshan/DGAE.

Multiview clustering as an important unsupervised method has been gathering a great deal of attention. However, most multiview clustering methods exploit the self-representation property to capture the relationship among data, resulting in high computation cost in calculating the self-representation coefficients. In addition, they usually employ different regularizers to learn the representation tensor or matrix from which a transition probability matrix is constructed in a separate step, such as the one proposed by Wu et al.. Thus, an optimal transition probability matrix cannot be guaranteed. To solve these issues, we propose a unified model for multiview spectral clustering by directly learning an adaptive transition probability matrix (MCA2M), rather than an individual representation matrix of each view. Different from the one proposed by Wu et al., MCA2M utilizes the one-step strategy to directly learn the transition probability matrix under the robust principal component analysis framework. Unlike existing methods using the absolute symmetrization operation to guarantee the nonnegativity and symmetry of the affinity matrix, the transition probability matrix learned from MCA2M is nonnegative and symmetric without any postprocessing. An alternating optimization algorithm is designed based on the efficient alternating direction method of multipliers. Extensive experiments on several real-world databases demonstrate that the proposed method outperforms the state-of-the-art methods.

Ball milling is a widely used method to produce graphene and other two-dimensional (2D) materials for both industry and research. Conventional ball milling generates strong impact forces, producing small and thick nanosheets that limit their applications. In this study, a viscous solvent-assisted planetary ball milling method has been developed to produce large thin 2D nanosheets. The viscous solvent simultaneously increases the exfoliation energy (Ee) and lowers the impact energy (Ei). Simulations show a giant ratio of η = Ee/Ei, for the viscous solvent, 2 orders of magnitude larger than that of water. The method provides both a high exfoliation yield of 74%, a high aspect ratio of the generated nanosheets of 571, and a high quality for a representative 2D material of boron nitride nanosheets (BNNSs). The large thin BNNSs can be assembled into high-performance functional films, such as separation membranes and thermally conductive flexible films with some performance parameters better than those 2D nanosheets produced by chemical exfoliation methods.

Real photograph denoising is extremely challenging in low-level computer vision since the noise is sophisticated and cannot be fully modeled by explicit distributions. Although deep-learning techniques have been actively explored for this issue and achieved convincing results, most of the networks may cause vanishing or exploding gradients, and usually entail more time and memory to obtain a remarkable performance. This article overcomes these challenges and presents a novel network, namely, PID controller guide attention neural network (PAN-Net), taking advantage of both the proportional-integral-derivative (PID) controller and attention neural network for real photograph denoising. First, a PID-attention network (PID-AN) is built to learn and exploit discriminative image features. Meanwhile, we devise a dynamic learning scheme by linking the neural network and control action, which significantly improves the robustness and adaptability of PID-AN. Second, we explore both the residual structure and share-source skip connections to stack the PID-ANs. Such a framework provides a flexible way to feature residual learning, enabling us to facilitate the network training and boost the denoising performance. Extensive experiments show that our PAN-Net achieves superior denoising results against the state-of-the-art in terms of image quality and efficiency. [ABSTRACT FROM AUTHOR]

In response to the demand for customs detection and tracking of individuals carrying radioactive sources, this paper proposes a method that integrates radiographic imaging information for identifying and tracking carriers. Initially, a portable source personnel visual tracking system equipped with a gamma camera is constructed to capture the position of radioactive sources and images of moving personnel within the security inspection area. Subsequently, a method is developed to link personnel targets identified by both the gamma camera and surveillance cameras, using Euclidean distance and intersecting areas to pinpoint carriers. Furthermore, leveraging the improved RepVGG reparameterization structure and residual networks, we accelerate Re-Identification (RE-ID) inference and propose a Deep Feature Tracking (DFT) method. This method tracks personnel by comparing feature similarities in adjacent frames. Experimental results validate the effectiveness of this method in identifying suspicious individuals carrying radioactive sources and in mapping the walking trajectories of these carriers using pre-deployed surveillance cameras.

Designing high-dimensional chaotic maps with expected dynamic properties is an attractive but challenging task. The dynamic properties of a chaotic system can be reflected by the Lyapunov exponents (LEs). Using the inherent relationship between the parameters of a chaotic map and its LEs, this paper proposes an $n$ -dimensional polynomial chaotic system ($n\text{D}$ -PCS) that can generate $n\text{D}$ chaotic maps with any desired LEs. The $n\text{D}$ -PCS is constructed from $n$ parametric polynomials with arbitrary orders, and its parameter matrix is configured using the preliminaries in linear algebra. Theoretical analysis proves that the $n\text{D}$ -PCS can produce high-dimensional chaotic maps with any desired LEs. To show the effects of the $n\text{D}$ -PCS, two high-dimensional chaotic maps with hyperchaotic behaviors were generated. A microcontroller-based hardware platform was developed to implement the two chaotic maps, and the test results demonstrated the randomness properties of their chaotic signals. Performance evaluations indicate that the high-dimensional chaotic maps generated from $n\text{D}$ -PCS have the desired LEs and more complicated dynamic behaviors compared with other high-dimensional chaotic maps. In addition, to demonstrate the applications of $n\text{D}$ -PCS, we developed a chaos-based secure communication scheme. Simulation results show that $n\text{D}$ -PCS has a stronger ability to resist channel noise than other high-dimensional chaotic maps. [ABSTRACT FROM AUTHOR]

The existing models of reversible data hiding in encrypted images (RDH-EI) are based on single data-hider, where the original image cannot be reconstructed when the data-hider is damaged. To address this issue, this article proposes a novel model with multiple data-hiders for RDH-EI based on secret sharing. It divides the original image into multiple different encrypted images with the same size of the original image and distributes them to multiple different data-hiders for data hiding. Each data-hider can independently embed data into the encrypted image to obtain the corresponding marked encrypted image. The original image can be losslessly recovered by collecting sufficient marked encrypted images from undamaged data-hiders when individual data-hiders are subjected to potential damage. This further protects the security of the original image. We provide four cases of the proposed model, namely, two joint cases and two separable cases. From the proposed model, we derive a separable RDH-EI method with high-capacity. Experimental results are presented to illustrate the effectiveness of the proposed method.

The characteristics of contact interfaces such as the distribution uniformity of the contact pressure and the effective contact area play a crucial role in engineering equipment. To investigate the influences of the variable material stiffness optimization (VMSO) design on the contact characteristics of the contact interfaces in an assembly, an heuristic-based VMSO algorithm is developed in this paper. A bi-objective function is defined by including both the distribution uniformity of the contact pressure and the effective contact area. A single bolted joint model is adopted as a design example. The results indicate that optimizing the stiffness of the materials around the contact interface is an effective approach to enhance the distribution uniformity of the contact pressure, increase the effective contact area and decrease the maximum contact pressure. Furthermore, the improvement effectiveness provided by the proposed variable stiffness design is better than that provided by the traditional variable thickness design.

Chaotic systems are an effective tool for various applications, including information security and internet of things. Many chaotic systems may have the weaknesses of incomplete output distributions, discontinuous chaotic regions, and simple chaotic behaviors. These may result in many negative influences in practical applications utilizing chaos. To deal with these issues, this study introduces a modular chaotification model (MCM) to increase the dynamic properties of current one-dimensional (1D) chaotic maps. To exhibit the effect of the MCM, three 1D chaotic maps are improved using the MCM as examples. Studies of the resulting properties show the robust and complex dynamics of these improved chaotic maps. Moreover, we implement these improved chaotic maps of MCM in a field-programmable gate array hardware platform and apply them to the application of PRNG. Performance analyses verify that these chaotic maps improved by the MCM have more complicated chaotic behaviors and wider chaotic ranges than the existing and several new chaotic maps.

To learn the self-representation matrices/tensor that encodes the intrinsic structure of the data, existing multiview self-representation models consider only the multiview features and, thus, impose equal membership preference across samples. However, this is inappropriate in real scenarios since the prior knowledge, e.g., explicit labels, semantic similarities, and weak-domain cues, can provide useful insights into the underlying relationship of samples. Based on this observation, this article proposes a prior knowledge regularized multiview self-representation (P-MVSR) model, in which the prior knowledge, multiview features, and high-order cross-view correlation are jointly considered to obtain an accurate self-representation tensor. The general concept of “prior knowledge” is defined as the complement of multiview features, and the core of P-MVSR is to take advantage of the membership preference, which is derived from the prior knowledge, to purify and refine the discovered membership of the data. Moreover, P-MVSR adopts the same optimization procedure to handle different prior knowledge and, thus, provides a unified framework for weakly supervised clustering and semisupervised classification. Extensive experiments on real-world databases demonstrate the effectiveness of the proposed P-MVSR model. [ABSTRACT FROM AUTHOR]

Vehicular Ad-hoc Networks (VANETs) provide key support for the achievement of intelligent, safe, and efficient driverless transportation systems through real-time communication between vehicles and vehicles, and vehicles and road infrastructure. This paper investigates a joint optimization problem of electric vehicles (EVs) charging management and resource allocation based on VANETs. EV charging requires significantly more time than refueling conventional vehicles, a key factor behind people's reluctance to transition from internal combustion engine vehicles to EVs. Previous works have primarily concentrated on fully-charged vehicles and random matching, which does not solve the problems of vehicle charging delays and long customer waiting times. Considering these factors, we proposes a distributed multi-level charging strategy and level-by-level matching method. Specifically, EVs and passengers are categorized into classes based on battery power and target mileage. Vehicles are then allocated to customers in the same or lower levels. Furthermore, the Attentive Temporal Convolutional Networks-Long Short Term Memory (ATCN-LSTM) model is leveraged to predict historical traffic data, supporting anticipatory decision-making. Subsequently, we develop a hierarchical charging and rebalancing joint optimization framework that incorporates charging facility planning. Experimental results obtained under various model parameters exhibit the method's commendable performance, as evidenced by metrics such as operating cost, system response time, and vehicle utilization.

Benefited from quaternion representation that is able to encode the cross-channel correlation of color images, quaternion principle component analysis (QPCA) was proposed to extract features from color images while reducing the feature dimension. A quaternion covariance matrix (QCM) of input samples was constructed, and its eigenvectors were derived to find the solution of QPCA. However, eigen-decomposition leads to the fixed solution for the same input. This solution is susceptible to outliers and cannot be further optimized. To solve this problem, this paper proposes a novel quaternion ridge regression (QRR) model for two-dimensional QPCA (2D-QPCA). We mathematically prove that this QRR model is equivalent to the QCM model of 2D-QPCA. The QRR model is a general framework and is flexible to combine 2D-QPCA with other technologies or constraints to adapt different requirements of real-world applications. Including sparsity constraints, we then propose a quaternion sparse regression model for 2D-QSPCA to improve its robustness for classification. An alternating minimization algorithm is developed to iteratively learn the solution of 2D-QSPCA in the equivalent complex domain. In addition, 2D-QPCA and 2D-QSPCA can preserve the spatial structure of color images and have a low computation cost. Experiments on several challenging databases demonstrate that 2D-QPCA and 2D-QSPCA are effective in color face recognition, and 2D-QSPCA outperforms the state of the arts. [ABSTRACT FROM AUTHOR]

Abstract Chaos is known as a natural candidate for cryptography applications owing to its properties such as unpredictability and initial state sensitivity. However, certain chaos-based cryptosystems have been proven to exhibit various security defects because their used chaotic maps do not have complex dynamical behaviors. To address this problem, this paper introduces a cosine-transform-based chaotic system (CTBCS). Using two chaotic maps as seed maps, the CTBCS can produce chaotic maps with complex dynamical behaviors. For illustration, we produce three chaotic maps using the CTBCS and analyze their chaos complexity. Using one of the generated chaotic maps, we further propose an image encryption scheme. The encryption scheme uses high-efficiency scrambling to separate adjacent pixels and employs random order substitution to spread a small change in the plain-image to all pixels of the cipher-image. The performance evaluation demonstrates that the chaotic maps generated by the CTBCS exhibit substantially more complicated chaotic behaviors than the existing ones. The simulation results indicate the reliability of the proposed image encryption scheme. Moreover, the security analysis demonstrates that the proposed image encryption scheme provides a higher level of security than several advanced image encryption schemes. [ABSTRACT FROM AUTHOR]

Most learning methods contain optimization as a substep, where the nondifferentiability and multimodality of objectives push forward the interplay of evolutionary optimization algorithms and machine learning models. The recently emerged evolutionary multimodal optimization (MMOP) technique enables the learning of diverse sets of effective parameters for the models simultaneously, providing new opportunities to the applications requiring both accuracy and diversity, such as ensemble, interactive, and interpretive learning. Targeting at locating multiple optima simultaneously in the multimodal landscape, this paper develops an efficient neighborhood-based niching algorithm. Bare-bones differential evolution is used as the baseline. Further, using Gaussian mutation with local mean and standard deviations, the neighborhoods capture niches that match well with the contours of peaks in the landscape. To increase diversity and enhance global exploration, the proposed algorithm embeds a diversity preserving operator to reinitialize converged or overlapped neighborhoods. The experimental results verify that the proposed algorithm has superior and consistent performance for a wide range of MMOP problems. Further, the algorithm has been successfully applied to train neural network ensembles, which validates its effectiveness and benefits of learning multimodal parameters. [ABSTRACT FROM AUTHOR]

We first verify the critical role of solvent evaporation on the resolution of inkjet printing. To confirm our hypothesis, we adjusted the evaporation rate gradient along the surface of adjacent droplets by controlling the drying microenvironment. Uneven solvent evaporation flux caused thermocapillary surface flow inward the space of micrometer-sized droplets and increase the air pressure, which prevented the neighboring droplets from coalescence. When reducing the droplet distance by the solvent evaporation-based method, a uniform profile could be obtained at the same time. This work brings us a step closer to resolving one of the critical bottlenecks to commercializing printed electronic goods.

Motivated by the concept of circuit design in digital circuit, this paper proposes a one-dimensional (1D) nonlinear model (1D-NLM) for producing 1D discrete-time chaotic maps. Our previous works have designed four nonlinear operations of generating new chaotic maps. However, they focus only on discussing individual nonlinear operations and their properties, but fail to consider their relationship among these operations. The proposed 1D-NLM includes these existing nonlinear operations, develops two new nonlinear operations, discusses their relationship among different nonlinear operations, and investigates the properties of different combinations of these operations. To show the effectiveness of 1D-NLM in generating new chaotic maps, as examples, we provide four new chaotic maps and study their dynamics properties from following three aspects: equilibrium point, stability, and bifurcation diagram. Performance evaluations are provided using the Lyapunov exponent, Shannon entropy, correlation dimension, and initial state sensitivity. The evaluation results show that these new chaotic maps have more complex chaotic behaviors than existing ones. To demonstrate the performance of 1D-NLM in practical applications, we use a pseudo-random number generator (PRNG) to compare new and existing chaotic maps. The randomness test results indicate that new chaotic map generated by 1D-NLM shows better performance than existing ones in designing PRNG. [ABSTRACT FROM PUBLISHER]

The operation of image filtering is widely used to deblur digital images. However, using inappropriate masks, it can also blur images. Motivated by this concept, this paper introduces an image cipher using block-based scrambling and image filtering (IC-BSIF). To the best of our knowledge, this is the first time that image filtering has been used for image encryption. IC-BSIF uses the well-known substitution-permutation network and strictly follows the concepts of confusion and diffusion. The block-based scrambling is able to separate neighboring pixels to different rows and columns, and thus can efficiently weaken the strong correlations between adjacent pixels. Using randomly generated masks by the secret key, the image filtering can spread little change of plain-images to the entire pixels of cipher-images. Simulation results show that IC-BSIF can encrypt different kinds of images into noise-like ones, and security evaluations demonstrate that it can achieve better performance than several state-of-the-art encryption schemes. [ABSTRACT FROM AUTHOR]

Obtaining homogeneous surface profiles of inkjet-printed silver nanoparticle films (SNFs) is severely hampered by the classical coffee-ringphenomenon. Traditional approaches to suppress the coffee-ring effect are achieved by improving the uniformity of liquid evaporation rate or introducing inward Marangoni force during ink drying. However, these existing methods involve extra chemicals, treatments, or equipment that will definitely increase the production cost or reduce conductivity. In this study, we demonstrate an inexpensive and efficient method to obtain uniform surface profiles of inkjet-printed SNFs by rationally regulating the drying microenvironment. To this end, a number of surface profiles of printed dots and lines, such as concave, convex, and flat, were first obtained via this method. The underlying principle of this method was then investigated by analytical calculations based on the simplified vapor diffusion model. Consequently, homogeneous surface profiles of inkjet-printed SNFs were achieved on the basis of aforementioned analytical calculations. Our work provides a new possibility to regulate the surface profiles of inkjet-printed nanoparticle-based patterns in the form of low cost and efficiency.

The regularized kernel method for the ranking problem has attracted increasing attentions in machine learning. The previous regularized ranking algorithms are usually based on reproducing kernel Hilbert spaces with a single kernel. In this paper, we go beyond this framework by investigating the generalization performance of the regularized ranking with multiscale kernels. A novel ranking algorithm with multiscale kernels is proposed and its representer theorem is proved. We establish the upper bound of the generalization error in terms of the complexity of hypothesis spaces. It shows that the multiscale ranking algorithm can achieve satisfactory learning rates under mild conditions. Experiments demonstrate the effectiveness of the proposed method for drug discovery and recommendation tasks. [ABSTRACT FROM AUTHOR]

With complex properties of ergodicity, unpredictability and sensitivity to initial states, chaotic systems are widely used in cryptography. This paper proposes a two-dimensional Logistic-adjusted-Sine map (2D-LASM). Performance evaluations show that it has better ergodicity and unpredictability, and a wider chaotic range than many existing chaotic maps. Using the proposed map, this paper further designs a 2D-LASM-based image encryption scheme (LAS-IES). The principle of diffusion and confusion are strictly fulfilled, and a mechanism of adding random values to plain-image is designed to enhance the security level of cipher-image. Simulation results and security analysis show that LAS-IES can efficiently encrypt different kinds of images into random-like ones that have strong ability of resisting various security attacks. [ABSTRACT FROM AUTHOR]

Sudoku refers to a two-dimensional (2D) number-placement puzzle with simple constraints that there are no repeated digits in each row, each column, or each block. Motivated by this Sudoku configuration, we introduce a number of Sudoku associated matrix element representations besides the conventional representation using matrix row-column pair. Specifically, they are representations via Sudoku matrix row-digit pair, digit-row pair, column-digit pair, digit-column pair, block-digit pair, and digit-block pair. This means we can secretly represent matrix elements via a Sudoku matrix, and furthermore develop new Sudoku associated 2D parametric bijections. To demonstrate the effectiveness and randomness of bijections, we introduce a simple but effective Sudoku Associated Image Scrambler only using 2D Sudoku associated bijections for image scrambling without bandwidth expansion. Simulations and comparisons demonstrate that the proposed method outperforms several state-of-the-art methods. [ABSTRACT FROM AUTHOR]

To protect image contents, most existing encryption algorithms are designed to transform an original image into a texture-like or noise-like image which is, however, an obvious visual sign indicating the presence of an encrypted image and thus results in a significantly large number of attacks. To address this problem, this paper proposes a new image encryption concept to transform an original image into a visually meaningful encrypted one. As an example of the implementation of this concept, we introduce an image encryption system. Simulation results and security analysis demonstrate excellent encryption performance of the proposed concept and system. [ABSTRACT FROM AUTHOR]

Because of the excellent properties of unpredictability, ergodicity and sensitivity to their parameters and initial values, chaotic maps are widely used in security applications. In this paper, we introduce a new two-dimensional Sine Logistic modulation map (2D-SLMM) which is derived from the Logistic and Sine maps. Compared with existing chaotic maps, it has the wider chaotic range, better ergodicity, hyperchaotic property and relatively low implementation cost. To investigate its applications, we propose a chaotic magic transform (CMT) to efficiently change the image pixel positions. Combining 2D-SLMM with CMT, we further introduce a new image encryption algorithm. Simulation results and security analysis demonstrate that the proposed algorithm is able to protect images with low time complexity and a high security level as well as to resist various attacks. [ABSTRACT FROM AUTHOR]