Representation-based Methods Research Articles

SAR target recognition through adaptive kernel sparse representation model based on local contrast perception

Sparse representation-based methods have shown promising prospects in synthetic aperture radar (SAR) target recognition due to their robustness under complex application conditions. However, these methods are still susceptible to the feature representation of SAR images. In this paper, we propose a new SAR target recognition method using an adaptive kernel sparse representation model based on local contrast perception (LCP). The LCP method combines a specially designed random projection matrix with two-dimensional filtering to capture extensive intensity contrasts between local regions in SAR images. A response map reconstruction technique shows that LCP can obtain discriminative properties between target and background regions and improve feature representation. The LCP features are then transformed into the kernel representation to improve linear separability by defining a reproducing kernel Hilbert space (RKHS) and using a nonlinear kernel function. The kernel representation is utilized to construct a new kernel sparse representation model for classification. Furthermore, neighborhood analysis (NA) is proposed to establish the relationship between the parameter and feature distribution, with an adaptive regularization parameter selection strategy to mitigate the parameter sensitivity. Experiments conducted on the moving and stationary target acquisition and recognition dataset demonstrate the superiority and robustness of the proposed method under different application conditions.

Hyperspectral anomaly detection based on adaptive background dictionary construction and collaborative representation

ABSTRACT Collaborative representation-based (CR) methods have received widespread attention in hyperspectral anomaly detection, but the results are greatly affected by the quality of the background dictionary. Abnormal pixels and abnormal-mixed pixels in the background dictionary may affect the accuracy of linear representation and make its performance poor. To address the above problem, an adaptive background dictionary construction-based anomaly detection method is proposed. To ensure the purity of the background dictionary, abnormal pixels and abnormal-mixed pixels around each test pixel are removed adaptively through clustering and pure pixel extraction. Furthermore, the saliency weight of the test pixel is calculated through the pixels in the inner window and weighted into the linear representation process to improve the robustness of the method. Experimental results on four hyperspectral datasets show that the proposed method performs better than other CR methods and traditional detectors, and it can reduce the dependence of anomaly detection performance on dual window size.

Multi-Modal Hierarchical Attention Model for Enhanced Phishing Website Detection

Phishing website attacks persist as a major cyber threat, continually evolving to evade detection. Existing detection methods, such as lookup systems and fraud cue-based approaches, have limitations, prompting the need for advanced techniques capable of addressing these challenges. This paper introduces a novel approach, the Multi-Modal Hierarchical Attention Model (MMHAM), designed to enhance phishing website detection by jointly analysing three crucial modalities: URLs, textual information, and visual design. Traditional lookup systems often fall short in addressing newly created attacks, while fraud cue-based methods may rely heavily on feature engineering, limiting their effectiveness. Deep representation-based methods have shown promise in learning intricate fraud cues, primarily focusing on URLs. However, they neglect the analysis of textual content and visual design, two equally important aspects of website content. MMHAM integrates information from URLs, textual content, and visual design through a shared dictionary learning approach. This innovative mechanism aligns representations from different modalities within the attention model, enabling the model to learn deep fraud cues comprehensively. The proposed MMHAM not only improves phishing detection capabilities but also introduces a hierarchical interpretability system. This system enhances model trustworthiness and provides actionable intelligence for informed decision-making at various levels of phishing threat detection. In our evaluation experiments, MMHAM demonstrated superior performance compared to existing methods, showcasing its ability to learn enhanced deep cues for phishing detection. Furthermore, the hierarchical interpretability system enabled the extraction of valuable phishing threat intelligence. This intelligence can be leveraged to inform phishing website detection strategies at different levels, empowering cybersecurity professionals with a more robust and adaptable defines against phishing attacks. The MMHAM model represents a significant step forward in the ongoing effort to combat the evolving landscape of cyber threats, particularly in the realm of phishing website detection.

Density-Based Discriminative Nonnegative Representation Model for Imbalanced Classification

Representation-based methods have found widespread applications in various classification tasks. However, these methods cannot deal effectively with imbalanced data scenarios. They tend to neglect the importance of minority samples, resulting in bias toward the majority class. To address this limitation, we propose a density-based discriminative nonnegative representation approach for imbalanced classification tasks. First, a new class-specific regularization term is incorporated into the framework of a nonnegative representation based classifier (NRC) to reduce the correlation between classes and improve the discrimination ability of the NRC. Second, a weight matrix is generated based on the hybrid density information of each sample’s neighbors and the decision boundary, which can assign larger weights to minority samples and thus reduce the preference for the majority class. Furthermore, the resulting model can be efficiently optimized through the alternating direction method of multipliers. Extensive experimental results demonstrate that our proposed method is superior to numerous state-of-the-art imbalanced learning methods.

REB: Reducing biases in representation for industrial anomaly detection

Existing representation-based methods usually conduct industrial anomaly detection in two stages: obtain feature representations with a pre-trained model and perform distance measures for anomaly detection. Among them, K-nearest neighbor (KNN) retrieval-based anomaly detection methods show promising results. However, the features are not fully exploited as these methods ignore domain bias of pre-trained models and the difference of local density in feature space, which limits the detection performance. In this paper, we propose Reducing Biases (REB) in representation by considering the domain bias and building a self-supervised learning task for better domain adaption with a defect generation strategy (DefectMaker) that ensures a strong diversity in the synthetic defects. Additionally, we propose a local-density KNN (LDKNN) to reduce the local density bias in the feature space and obtain effective anomaly detection. The proposed REB method achieves a promising result of 99.5% Im.AUROC on the widely used MVTec AD, with smaller backbone networks such as Vgg11 and Resnet18. The method also achieves an impressive 88.8% Im.AUROC on the MVTec LOCO AD dataset and a remarkable 96.0% on the BTAD dataset, outperforming other representation-based approaches. These results indicate the effectiveness and efficiency of REB for practical industrial applications. Code: https://github.com/ShuaiLYU/REB.

Unsupervised anomaly detection in images using attentional normalizing flows

Image-based anomaly detection has been widely used in practice, but it is still a challenging task due to the irregularity of anomalies. Existing representation-based methods have achieved high accuracy metrics in image-based anomaly detection, but they are weak in capturing anomalous regions, resulting in small inter-class variance between the latent distributions of normal and anomalous data. Furthermore, most current neural networks have low parameter efficiency, which limits their inference performance. To address these issues, we propose a novel normalizing flow called AFlow. Our network represents multi-scale features with multi-scale convolutions instead of inputs of different resolutions, significantly reducing the network parameters and improving the fine-grained representation in features. Then, an attention module is introduced into the normalizing flow to further select the multi-scale features along feature channels, making the network focus on relevant regions with more object details to refine the objective of maximum likelihood estimation. Experiments on two public datasets show that the proposed method can achieve more accurate prediction results and higher parameter efficiency than previous detection models: the average AUROC of the proposed method reaches 99.75% and 97.2% on MVTec AD and BTAD datasets, respectively.

Hyperspectral Image Super-Resolution via Adaptive Factor Group Sparsity Regularization-Based Subspace Representation

Hyperspectral image (HSI) super-resolution is a vital technique that generates high spatial-resolution HSI (HR-HSI) by integrating information from low spatial-resolution HSI with high spatial-resolution multispectral image (MSI). However, existing subspace representation-based methods face challenges, including adaptive subspace dimension determination, inadequate spectral correlation capture, and expensive computation. In this paper, we propose a novel factor group sparsity regularized subspace representation (FGSSR)-based method for HSI super-resolution that can simultaneously address these issues encountered in previous methods. Specifically, by incorporating the factor group sparsity regularization into the subspace representation model, we first propose an FGSSR model to capture the spectral correlation property of the HR-HSI. The key advantage of FGSSR lies in its equivalence to the Schatten-p norm and its adaptive determination of the accurate subspace dimension, enabling it to capture spectral correlation more effectively. To preserve the spatial self-similarity prior in the HR-HSI, the tensor nuclear norm regularization on the low-dimensional coefficients is also incorporated into the proposed FGSSR-based model. Finally, an effective proximal alternating minimization-based algorithm is developed to solve the FGSSR-based model. Experimental results on the simulated and real datasets demonstrate that the proposed FGSSR-based method outperforms several state-of-the-art fusion methods with significant improvements.

Target Detection in Hyperspectral Remote Sensing Image: Current Status and Challenges

Abundant spectral information endows unique advantages of hyperspectral remote sensing images in target location and recognition. Target detection techniques locate materials or objects of interest from hyperspectral images with given prior target spectra, and have been widely used in military, mineral exploration, ecological protection, etc. However, hyperspectral target detection is a challenging task due to high-dimension data, spectral changes, spectral mixing, and so on. To this end, many methods based on optimization and machine learning have been proposed in the past decades. In this paper, we review the representatives of hyperspectral image target detection methods and group them into seven categories: hypothesis testing-based methods, spectral angle-based methods, signal decomposition-based methods, constrained energy minimization (CEM)-based methods, kernel-based methods, sparse representation-based methods, and deep learning-based methods. We then comprehensively summarize their basic principles, classical algorithms, advantages, limitations, and connections. Meanwhile, we give critical comparisons of the methods on the summarized datasets and evaluation metrics. Furthermore, the future challenges and directions in the area are analyzed.

Impact of Amplitude and Phase of fMRI time series for Functional Connectivity Analysis

PurposeSeveral studies in the field of fMRI have reported the synchrony between the brain regions using instantaneous phase (IP) representation (derived from analytic representation of BOLD time series). We hypothesized that instantaneous amplitude (IA) representation from different brain regions might give additional information to the functional brain networks. To validate this, we explored this representation of resting state BOLD fMRI signal for deriving resting state networks (RSNs) and compared it with the IP representation based RSNs. MethodResting state fMRI data of 100 healthy adults (age=20–35 years, 54 females) from the population of 500 Subjects of HCP dataset were studied. Data was acquired using a 3T scanner in four runs (15-min each) with the phase encoding directions: Left to Right (LR), Right to Left (RL). These four runs were acquired in two sessions, and subjects were asked to keep their eyes open with a fixation on a white cross. The IA and IP representations were derived from a narrow-band filtered BOLD time series using Hilbert transforms and a seed-based approach is used to compute the RSNs in the brain. ResultsThe experimental results demonstrate that within the frequency range 0.01–0.1 Hz, IA representation based RSNs have the highest similarity score between the two sessions for the motor network. Whereas for fronto-parietal network, IP based activation maps have the highest similarity score for all the frequency bands. For higher frequency band (0.198–0.25 Hz) consistency of the obtained RSNs across two sessions reduced for both IA and IP representations. Fusion of IA and IP representations based RSNs in comparison to those of IP based representation, leads to 3–10% improvement in the similarity scores between the default mode network obtained for the two sessions. In addition, the same comparison demonstrates 15–20% improvement for the motor network in the frequency bands: 0.01–0.04 Hz, 0.04–0.07 Hz, slow5 (0.01–0.027 Hz) and slow-4 (0.027–0.073 Hz). It is also observed that the similarity score between two sessions using instantaneous frequency (IF) (derivative of unwrapped IP) representation in exploring functional connectivity (FC) networks is comparable with those obtained using IP representation. ConclusionOur findings suggest that IA-representation based measures can estimate RSNs with the reproducibility between the sessions comparable to that of the IP representation-based methods. This study demonstrates that IA and IP representations contain the complementary information of BOLD signal, and their fusion improves the results of FC.

Trust-worth multi-representation learning for audio classification with uncertainty estimation

Multi-view learning has been explored for audio classification tasks, exploiting different representations of audio signals, ranging from MFCC, CQT, to raw signals. The quality of each view may vary for different audio signals, and the appropriate uncertainty quantification for each view has not been fully explored. In this work, we explore a trusted multi-view learning framework for classification tasks in order to fully incorporate different views. Our framework consists of three parallel branches of Transformer architectures (Gammatone spectrogram, log-Mel and CQT) and they are combined using the uncertainty estimation of different branch. In addition to computing the classification probabilities, the uncertainty of each representation can also be obtained using the framework. We firstly calculate the evidence based on feature vectors to obtain the probabilities and the uncertainty of classification problems for Gammatone, log-Mel and CQT branch. By integrating the confidence from each of the different representations using the Dempster–Shafer theory, the classification framework can provide higher accuracy and confidence. To demonstrate the effectiveness of the proposed framework, we conduct the experiments on the GTZAN dataset. The obtained results show that our method can reach the accuracy of 83.0%, which significantly outperforms single representation-based methods while providing uncertainty estimation for different views.

PAEDID: Patch Autoencoder-based Deep Image Decomposition for pixel-level defective region segmentation

Unsupervised pixel-level defective region segmentation is an important task in image-based anomaly detection for various industrial applications. The state-of-the-art methods have their own advantages and limitations: matrix-decomposition-based methods are robust to noise, but lack complex background image modeling capability; representation-based methods are good at defective region localization, but lack accuracy in defective region shape contour extraction; reconstruction-based methods detected defective region match well with the ground truth defective region shape contour, but are noisy. To combine the best of both worlds, we present an unsupervised Patch AutoEncoder-based Deep Image Decomposition (PAEDID) method for defective region segmentation. In the training stage, we learn the common background as a deep image prior by a patch autoencoder network. In the inference stage, we formulate anomaly detection as an image decomposition problem with the deep image prior and sparsity regularizations. By adopting the proposed approach, the defective regions in the image can be accurately extracted in an unsupervised fashion. We demonstrate the effectiveness of the PAEDID method in simulation studies and an industrial dataset in the case study.

Multiscale Multidirection Binary Pattern Learning for Discriminant Palmprint Identification

Binary pattern learning has been achieved promising recognition performance in palmprint identification. However, most of direction representation-based methods are hand-crafted. They present the direction characteristics only in one scale, and thus the comprehensive information from multiple scales and multiple directions are ignored. This is also true especially for the binary coding-based methods. As a solution to the problem, this paper proposes a multi-scale multi-direction binary (MSMDB) pattern learning method for palmprint identification. Specifically, we first form a feature container of multi-scales multi-directions for palmprint images. The container contains robust direction average vectors (DAVs). Then, a learning model is proposed to learn robust and discriminant direction information from the extracted convolution average features. The proposed model can project these DAVs into discriminant direction binary codes. The model not only maximizes the variance of the learned binary codes and inter-class distance, but also minimizes intra-class distance. Finally, we cluster all block histograms of multi-scale projections to form the discriminative direction binary palmprint descriptors for palmprint identification. We evaluate the performance of the proposed method on five public databases compared to other related methods. For example, MSMDB can achieve the best identification accuracy of 97.93 percent when the number of training data is 2 on the database IITD.

Joint learning model for underwater acoustic target recognition

In underwater acoustic target recognition, wave-based methods and time–frequency (T-F) representation-based methods are typically used to identify the target from different perspectives. Both methods have advantages and disadvantages. In this study, a complementary space between the wave-based model and the T-F representation-based model was proven to exist. Advanced lightweight technologies make the fusion of the two types of models technically possible. First, a lightweight multiscale residual deep neural network (MSRDN) is designed using lightweight network design techniques, in which 64.18% of parameters and 79.45% of floating point operations (FLOPs) are reduced from the original MSRDN with a small loss of accuracy. Then, a joint model combining wave and T-F representation-based models was presented. An effective synchronous deep mutual learning method that saves approximately 11.54%–16.27% training time is proposed to train the joint model. Two datasets acquired from real-world scenarios were used to verify the effectiveness of the proposed method. Compared with state-of-the-art methods, the joint model with synchronous deep mutual learning achieved the best recognition accuracies of 85.20% and 79.50% in the two datasets, respectively. The results of ablation explorations prove that the performance improvements of the proposed methods benefit from the deep mutual learning of the two branches and are not unique to certain models. Finally, a discussion reveals the essential mechanism of the proposed method.

Representation based regression for object distance estimation

In this study, we propose a novel approach to predict the distances of the detected objects in an observed scene. The proposed approach modifies the recently proposed Convolutional Support Estimator Networks (CSENs). CSENs are designed to compute a direct mapping for the Support Estimation (SE) task in a representation-based classification problem. We further propose and demonstrate that representation-based methods (sparse or collaborative representation) can be used in well-designed regression problems especially over scarce data. To the best of our knowledge, this is the first representation-based method proposed for performing a regression task by utilizing the modified CSENs; and hence, we name this novel approach as Representation-based Regression (RbR). The initial version of CSENs has a proxy mapping stage (i.e., a coarse estimation for the support set) that is required for the input. In this study, we improve the CSEN model by proposing Compressive Learning CSEN (CL-CSEN) that has the ability to jointly optimize the so-called proxy mapping stage along with convolutional layers. The experimental evaluations using the KITTI 3D Object Detection distance estimation dataset show that the proposed method can achieve a significantly improved distance estimation performance over all competing methods. Finally, the software implementations of the methods are publicly shared at https://github.com/meteahishali/CSENDistance.

Deep treatment-adaptive network for causal inference

Causal inference is capable of estimating the treatment effect (i.e., the causal effect of treatment on the outcome) to benefit the decision making in various domains. One fundamental challenge in this research is that the treatment assignment bias in observational data. To increase the validity of observational studies on causal inference, representation-based methods as the state-of-the-art have demonstrated the superior performance of treatment effect estimation. Most representation-based methods assume all observed covariates are pre-treatment (i.e., not affected by the treatment) and learn a balanced representation from these observed covariates for estimating treatment effect. Unfortunately, this assumption is often too strict a requirement in practice, as some covariates are changed by doing an intervention on treatment (i.e., post-treatment). By contrast, the balanced representation learned from unchanged covariates thus biases the treatment effect estimation. In light of this, we propose a deep treatment-adaptive architecture (DTANet) that can address the post-treatment covariates and provide a unbiased treatment effect estimation. Generally speaking, the contributions of this work are threefold. First, our theoretical results guarantee DTANet can identify treatment effect from observations. Second, we introduce a novel regularization of orthogonality projection to ensure that the learned confounding representation is invariant and not being contaminated by the treatment, meanwhile mediate variable representation is informative and discriminative for predicting the outcome. Finally, we build on the optimal transport and learn a treatment-invariant representation for the unobserved confounders to alleviate the confounding bias.

Non-parametric modelling and simulation of spatiotemporally varying geo-data

ABSTRACT Modelling and simulation of spatially or temporally varying geo-data play a pivotal role in the development of digital twins of civil infrastructures and smart cities. Measurements on geo-data are however often sparse, and it is challenging to model or simulate the spatiotemporally varying geo-data directly from sparse measurements. Non-parametric methods are appealing to tackle this challenge because they bypass the difficulty in the selection of suitable parametric models or function forms and offer great flexibility for mimicking complicated characteristics of geo-data in a data-driven manner. This paper provides a state-of-the-art review of non-parametric modelling and simulation of spatiotemporally varying geo-data under the framework of spectral representation or compressive sensing/sampling (CS). Similarity and differences between the spectral representation-based methods and the CS-based methods are discussed, including modelling of unknown trend function, marginal probability density function (PDF), and spatial or temporal autocovariance structure. Advantages of the CS-based methods are highlighted, such as superior performance for sparse measurements (i.e. capable of dealing with a sampling frequency lower than Nyquist frequency) and incorporation of the uncertainty associated with the interpretation of sparse measurements. Numerical examples are presented to demonstrate both spectral representation-based methods and CS-based methods.

Joint latent low-rank and non-negative induced sparse representation for face recognition

Representation-based methods have achieved exciting results in recent applications of face recognition. However, it is still challenging for the face recognition task due to noise and outliers in the data. Many existing methods avoid these problems by constructing an auxiliary dictionary from the extended data but fail to achieve good performances because they use the main dictionary only for classification. In this paper, to avoid the need to manually construct an auxiliary dictionary and the effects of noise, we propose a Joint Latent Low-Rank and Non-Negative Induced Sparse Representation (JLSRC) for face recognition. Specifically, JLSRC adaptively learns two clean low-rank reconstructed dictionaries jointly via an extended latent low-rank representation to reveal the potential relationships in the data and then embeds a non-negative constraint and an Elastic Net regularization in the coefficient vectors of the dictionaries to enhance the performance on classification. In this way, the learned low-rank dictionaries can be mutually boosted to extract discriminative features and handle the noise, and the obtained coefficient vectors are simultaneously both sparse and discriminative. Moreover, the proposed method seamlessly and elegantly integrates low-rank learning and sparse representation-based classification. Extensive experiments on three challenging face databases demonstrate the effectiveness and robustness of JLSRC in comparison with the state-of-the-art methods.

A Review of Vision-Based Techniques Applied to Detecting Human-Object Interactions in Still Images

Due to the rising demand for automatic interpretation of visual relationships in several domains, human-object interaction (HOI) detection and recognition have also gained more attention from researchers over the last decade. This survey paper concentrates on human-centric interactions, which can be categorized as human-to-human and human-to-objects. Although an extensive amount of research work has been done in this area, real-world constraints like the domain of possible interactions make the research a challenging task. This paper provides an analysis of conventional hand-crafted representation-based methods and recent deep learning-based methods, ongoing advancements taking place in the field of HOI recognition and detection, and challenges faced by the researchers. Moreover, we present a detailed picture of publicly available datasets for HOI evaluations. At the end, the future scope of the study is discussed.

Hyperspectral Image Classification via Multi-Feature-Based Correlation Adaptive Representation

In recent years, representation-based methods have attracted more attention in the hyperspectral image (HSI) classification. Among them, sparse representation-based classifier (SRC) and collaborative representation-based classifier (CRC) are the two representative methods. However, SRC only focuses on sparsity but ignores the data correlation information. While CRC encourages grouping correlated variables together but lacks the ability of variable selection. As a result, SRC and CRC are incapable of producing satisfied performance. To address these issues, in this work, a correlation adaptive representation (CAR) is proposed, enabling a CAR-based classifier (CARC). Specifically, the proposed CARC is able to explore sparsity and data correlation information jointly, generating a novel representation model that is adaptive to the structure of the dictionary. To further exploit the correlation between the test samples and the training samples effectively, a distance-weighted Tikhonov regularization is integrated into the proposed CARC. Furthermore, to handle the small training sample problem in the HSI classification, a multi-feature correlation adaptive representation-based classifier (MFCARC) and MFCARC with Tikhonov regularization (MFCART) are presented to improve the classification performance by exploring the complementary information across multiple features. The experimental results show the superiority of the proposed methods over state-of-the-art algorithms.

Hyperspectral Target Detection Method Based on Nonlocal Self-Similarity and Rank-1 Tensor

In recent years, many target detection methods based on tensor representation theory have been proposed and achieved good results for hyperspectral images (HSIs). However, these methods still have some deficiencies. For example, 3-D hyperspectral data are first transformed into 1-D vectors in these methods, which may destroy the spatial structure of HSI data and reduce the detection performance. Besides, when the number of training samples is small, the results of the target detection method usually become worse. To solve these problems, a hyperspectral target detection method based on nonlocal self-similarity and rank-1 tensor is proposed in this article. First, different from these traditional tensor representation-based methods, the third-order tensor data are directly used as the input of the proposed method to preserve the spatial information and structure of an HSI. Second, the tensor blocks related to the class are constructed by using the nonlocal self-similarity of HSI data. Finally, by taking advantage of rank-1 canonical decomposition attribute, the process of tensor operation can be simplified, and the number of training samples can be reduced. The proposed method is compared with six state-of-the-art hyperspectral target detection methods on four HSI data sets. The experimental results show that the proposed method can have better target detection results than other compared methods, especially in the case of fewer training samples.