Unsupervised visible-infrared person re-identification (USL-VI-ReID) aims to match person images across visible and infrared modalities without any labeled data, which is severely hindered by the large cross-modality discrepancy and the absence of ground-truth annotations. Recent advances predominantly adopt unsupervised contrastive learning frameworks that rely on clustering-generated pseudo-labels to guide representation learning. While existing methods emphasize establishing cross-modality correspondences for modality-invariant feature learning, they often overlook the adverse impact of unreliable pseudo-labels, which frequently arise from significant intra-class variations and inter-modality misalignment. Such noisy correspondences can severely degrade model robustness and generalization. To tackle this challenge, we propose Soft smooth Contrastive Learning with Hybrid Memory (SCLHM), a novel framework that jointly addresses noise pseudo-labels and cross-modality divergence. Specifically, we first design a Soft Smooth Contrastive Learning (SSCL) module that mitigates the influence of noise pseudo-labels by smoothing similarity distributions based on intra-class consistency. In addition, we introduce a Hybrid Memory Learning (HML) module that unifies modality-specific and modality-invariant feature representations, enabling more comprehensive knowledge integration. Furthermore, an Adaptive-weight Memory Update (AMU) strategy is developed to dynamically adjust memory bank updates during batch training, promoting the learning of globally discriminative and stable features. Code will be released.

Most Multi-view Graph-based Clustering (MGC) models always obtain suboptimal performance since the necessary symmetry of graph is ignored during the process of graph fusion. To solve the problem, we propose Multi-view Clustering based on Doubly Stochastic Graph (MCDSG). Our MCDSG precalculates Single-view Similarity Graphs (SSGs) and then fuses them into a consensus one with doubly stochastic (non-negative, sum-to-one and symmetry) constraints, directly providing clustering results by its connectivity. For optimization, a novel and easy-understanding Augmented Lagrangian Method (ALM) is proposed to substitute the widely used Von-Neumann Successive Projection (VNSP) method, which simultaneously optimizes all the doubly stochastic conditions to the optimal solution. To verify the robustness to noisy data sets, we propose a pipeline to add noise to the key features of face images and obtain a two-view data set termed NoisedORL. Experiments on both synthetic data sets and real benchmarks show that our MCDSG achieves SOTA clustering performance against nine methods. Code will be published at https://github.com/NianWang-HJJGCDX/MCDSG.

Recent weakly supervised image dehazing (WSID) works have succeeded to improve models’ generalization ability to real scene dehazing by using generative adversarial network (GAN) for unpaired image training. However, it is still difficult for current WSID methods to train one effective dehazing model for various scenes since 1) they always result in residual haze due to insufficient generalization to the feature distribution of real scenes, and 2) they are prone to cause distortions like color shifts, artifacts or halos etc, owing to embedding manual prior or threshold hypothesis for image reconstruction. To solve above problems, in this paper, we propose a novel WSID model via physics-based decomposition (PBD), which estimates atmospheric light, scattering coefficient and scene depth of real haze input to effectively capture the illumination information and haze distribution to recover a preliminary dehazed image by minimizing reconstruction loss. With this constraint, we subtly design a discrete wavelet discriminator (DWD) to effectively improve the generalization to real scene from both spatial and frequency aspect under the supervision of unpaired real clear image. Our PBD is a purely data-driven model freeing from any manual setting or partially correct prior, thus simultaneously ensuring the realness and visibility of dehazed images. Experiments on seven benchmarks verified the strong generalization ability of our PBD, which achieves SOTA dehazing performance with realistic details. Code will be published at https://github.com/NianWang-HJJGCDX/PBD.

Current multi-view graph-based methods often produce suboptimal outcomes due to unreliable input information and ineffective multi-view fusion. To overcome these limitations, we propose Dynamic and Consistent Doubly Stochastic (DCDS) similarity learning for multi-view and multi-order clustering. To improve input reliability, DCDS dynamically selects informative graphs and aggregates them into a 3-D tensor, thereby learning a consistent similarity in selected dimensions. For graph fusion, we introduce an efficient doubly stochastic graph learning strategy that unifies similarity integration and symmetry enforcement into a single step, explicitly revealing cluster structure without requiring postprocessing. DCDS formulates these components within a unified optimization framework to synergistically achieve optimal clustering performance. An Augmented Lagrange Multiplier (ALM) based method with a key equivalent transformation is proposed to solve the subproblem with doubly stochastic constraints. Extensive experiments demonstrate that DCDS learns graphs with better structural clarity, outperforming state-of-the-art methods and attaining ground-truth clustering results on the YaleA, MSRC, and HW2 datasets. Code will be published at https://github.com/NianWang-HJJGCDX/DCDS.git.

Most existing multi-view graph clustering (MGC) methods rely on spectral decomposition of n x n similarity matrices, suffering from O(n 3 ) computational complexity and inherent information loss due to continuous relaxation of discrete cluster indicators. To address these limitations, we propose HISSM, a novel discrete MGC framework that directly optimizes discrete cluster indicators without intermediate relaxations. Our approach introduces two key innovations: (1) a Supercluster Similarity Minimization (SSM) mechanism that jointly optimizes multi-view fusion and discrete clustering through an efficient objective, and (2) a hierarchical initialization (HI) strategy that guarantees stable initialization using only nearest-neighbor relationships. The complete framework achieves O(n 2 ) complexity with no hyperparameter tuning requirements. Extensive experiments demonstrate that HISSM achieves state-of-the-art clustering accuracy while being significantly faster than conventional methods. Beyond its immediate performance advantages, our work provides a new paradigm for discrete graph clustering that bypasses traditional spectral relaxation and matrix factorization constraints, opening new directions for effective and efficient clustering research. Code will be published at https://github.com/NianWang-HJJGCDX/HISSM.

Unsupervised visible-infrared person re-identification (USL-VI-ReID) presents significant challenges in cross-modality image matching without annotated training data. Recently, pseudo-label-based methods have emerged as a promising approach in USL-VI-ReID. However, existing methods ignore the discrepancy in the feature distribution caused by random data augmentation, resulting in noisy pseudo labels. Worse still, they usually explore the correspondence between the two modalities at modality-specific cluster-level, leading to insufficient cross-correspondences. In response, we propose a Modality-specific and Modality-agnostic Soft Contrastive Learning (MMSCL) method. Specifically, we design a modality-agnostic hybrid memory to initialize modality-shared prototypes to mitigate the impact of the cross-modality gap. In addition, we introduce soft contrastive learning with a mean-teacher framework to constrain the consistency of the feature distribution. Moreover, an Adaptive-weighted Memory Updating module is proposed to tackle the discrepancy distribution problem in the batch training process. Extensive experimental results on two benchmark datasets demonstrate that our proposed method achieves state-of-the-art performance.

Intelligent mapping technology has become more popular in the era of AI. However, traditional mapping accuracy evaluation methods, algebraic geometric methods (Hausdorff distance), and AI accuracy evaluation indicators based on IOU and its derivative methods, due to theoretical design flaws, are all unable to accurately calculate the true error values of the automated mapping results. Therefore, this paper proposes a theoretical accuracy evaluation method for intelligent mapping results based on automatic contour matching and Riemann integration. Firstly, we calculated the inflection-point matching relationship between the mapping contour and the real reference contour based on the proposed contour matching method. Next, using the inflection point matching relationship between the two contours, the two vector contours are divided into several groups of mutually matching edges. The distances between the matching edges were calculated using the proposed calculation method, and the weighted average distance of these distances based on the lengths of the matching edges was considered as the error of the mapping contour. Experiments were conducted on an ideal dataset and two groups of mapping-contour data generated by different automatic mapping algorithms. The experiments show that the accuracy evaluation effect of the proposed method is significantly better than that of the existing mapping accuracy evaluation methods.

Image dehazing is essential to boost the visual quality of images captured in hazy conditions. Recently, many learning-based methods were proposed to achieve single image dehazing with the training of tremendous paired synthetic hazy/ real clean images. Due to the domain gap between real and synthetic scenes, these models cannot generalize well to various real hazy scenes, leading to under-dehazed results. To overcome this problem, we propose a real scene image Dehazing Network with Multi-prior Guidance and Domain Transfer (DNMGDT). Our DNMGDT is based on a parameter shared architecture trained by synthetic hazy images and real hazy images simultaneously. For real hazy images, multiple prior-based dehazed images are adopted as pseudo clean images. An Image Quality Guided Adaptive Weighting (IQGAW) scheme is proposed to form the supervision by automatically weighting different parts of these prior-based dehazed images and suppressing negative information of them. Moreover, to reduce the domain gap between real and synthetic hazy scenes, a Physical Model Guided image level Domain Transfer (PMGDT) mechanism is proposed to regularize the learning process with consistency constraint. Experiments on various datasets demonstrated the effectiveness of our proposed method especially for real hazy scenes.

The automatic extraction of buildings from remote sensing images is crucial for various applications such as urban planning and management, emergency response, and map making and updating. In recent years, deep learning (DL) methods have made significant progress in this field. However, due to the complex and diverse structures of buildings and their interconnections, the accuracy of extracted buildings remains insufficient for high-precision applications such as maps and navigation. To address the issue of enhancing building boundary extraction, we propose a modified instance segmentation model, AP-PointRend (Adaptive Parameter-PointRend), to improve the performance of building instance extraction. Specifically, the model can adaptively select the number of iterations and points based on the size of buildings to improve the segmentation accuracy of large buildings. By introducing regularization constraints, discrete small patches are removed, preserving boundaries better during the segmentation process. We also designed an image merging method to eliminate seams, ensure the recall rate, and improve the extraction accuracy. The Vaihingen and WHU benchmark datasets were used to evaluate the performance of the AP-PointRend method. The experimental results showed that the proposed AP-PointRend approach generated better building extraction results compared with other state-of-the-art methods.

Hyperspectral image (HSI) clustering has drawn more and more attention in recent years as it frees us from labor-intensive manual annotation. However, current works cannot fully enjoy the rich spatial and spectral information due to redundant spectral signatures and fixed anchor learning. Moreover, the learned graph always obtains suboptimal results due to the separate affinity estimation and graph symmetry. To address the above challenges, in this paper, we propose large-scale hyperspectral image-projected clustering via doubly stochastic graph learning (HPCDL). Our HPCDL is a unified framework that learns a projected space to capture useful spectral information, simultaneously learning a pixel–anchor graph and an anchor–anchor graph. The doubly stochastic constraints are conducted to learn an anchor–anchor graph with strict probabilistic affinity, directly providing anchor cluster indicators via connectivity. Meanwhile, when using label propagation, pixel-level clustering results are obtained. An efficient optimization strategy is proposed to solve our HPCDL model, requiring monomial linear complexity concerning the number of pixels. Therefore, our HPCDL has the ability to deal with large-scale HSI datasets. Experiments on three datasets demonstrate the superiority of our HPCDL for both clustering performance and the time burden.

Diesel engine will emit a lot of high temperature flue gas to heat the exhaust pipe, which will radiate a lot of infrared signals to the outside world. To reduce the heat signal radiated by the exhaust pipe, the cold air film insulation protection method is proposed in this study. However, for the long exhaust pipe, the longer traveling distance makes the hot smoke rise phenomenon obvious, which forms a strong heating effect on the exhaust pipe. Therefore, this study proposes to increase the swirl effect on the cold air film and rely on rotation to increase the adhesion between the film and the tube wall. In this study, a numerical simulation method was used to compare the thermal protection effect of cold fluid on smoke exhaust pipe under swirling and non-swirling conditions with the numerical model verified by experiments. The temperature distribution, velocity distribution, carbon dioxide distribution and heat transfer to the pipe wall are discussed when the mixed jet with hot flue gas reaches steady state under swirling and non-swirling conditions. The results show that the cyclone cold fluid has better thermal protection, and the maximum temperature rise of the pre-swirling cold fluid is 90.39% lower than that of the non-swirling cold fluid. With the increase of cold fluid flow, the pipe temperature rise first decreases and then increases, and there exists a wall temperature rise that shows the best performance when the cold fluid flow is not the maximum.

Unsupervised video person re-identification (USL-VReID) aims to identify the same person across cameras without annotation. The key to this challenging task is to take advantage of the discriminative features of frames and tracklets. Most existing methods model tracklet features by mean average pooling and address the USL-VReID problem by using cluster centroid contrastive learning. However, the mean average pooling strategy and cluster centers primarily focus on commonality, overlooking divergencies. In this paper, we propose a Dual Representation Modeling and Progressive Contrastive Learning (DRMPCL) method to learn fine-grained features of the diversity distribution. Specifically, our approach explores the discriminant features of diversity distribution in both frame-level tracklet modeling and tracklet-level discrepant contrastive learning. Then a multi-level progressive learning strategy effectively optimizes the model from reliable to informative, and from commonality to divergence. Comprehensive experiments on two widely-used datasets demonstrate that DRMPCL outperforms the state-of-the-art methods.

Graph based clustering involves learning a proximity matrix with explicit clustering structure. However, since the limited link inputs and insufficient graph fusion, current works always obtain poor graphs with suboptimal clustering results. To solve the problem, in this paper, we propose a novel multi-order graph based clustering model via dynamic low-rank tensor approximation (MCDLT). Firstly, we use high-order proximity to enrich the link relations for graph inputs. Then a graph selection mechanism and low rank tensor approximation method are used to dynamically learn the consistent information from the complex links in various-order graphs. Finally, we propose a doubly stochastic graph fusion method to directly learn a symmetrical graph that provides clustering results by its connectivity. A novel Augmented Lagrangian Multiplier (ALM) based method is proposed for the sub-problem of doubly stochastic constraints. Experiments shows that our method learns a graph with clearer data structure, achieving SOTA clustering performance and obtaining the GroundTruth for JAFFE data set. Code will be published at https://github.com/NianWang-HJJGCDX/MCDLT.git.

The structured-3D-reconstruction of buildings is essential for rapid urban-3D-modeling. The existing data-driven methods for full-automatic building-contour-extraction fail to meet mapping-accuracy-requirements and lack robustness. Therefore, this paper proposes a novel data-driven algorithm for roof-contour-extraction and structured-3D-reconstruction by fusing stereo-images and LiDAR-point-clouds. Firstly, an improved Sobel-based-operator is designed, which effectively improves the detection rate of building-edge-features. Secondly, based on a buffer constraint strategy, a method for automatically removing non-building linear features that away from the building boundaries is proposed. Subsequently, to tackle the challenge of distinguishing pseudo-building linear features, an equidistant mapping method based on LiDAR profile and image features is presented, yielding high-confidence building-edge-features. Finally, to address the fragmentation and incompleteness issues on low-level image features, a set of rules for refining building-boundary-features and achieving complete contour closure is further proposed. Based on these, structured-3D-models are constructed quickly. Additionally, a vector-based evaluation metric RMEC is proposed, which allows for a better assessment of intelligent mapping. Experiments conducted on the ISPRS benchmark datasets indicate that the average correctness (CR) of the proposed method for extracting building-roof-contours is 96.3%, which is better than the other three methods; the average RMSE of the proposed method has a 12.3% improvement than the comparison method.

Graph-based clustering is a hot topic in machine learning, whose effectiveness highly relies on the quality of the learned graph. Recent researches preferred to learn the nearest doubly stochastic approximation of a graph to suppress intercluster connections and enhance intracluster connections and thus improve clustering performance. While current paradigm is limited by three key problems: 1) it is restricted by a predefined graph; 2) the separated stages of spectral decomposition-based way (graph learning, spectral embedding learning, and cluster assignment by k-means) cause mismatched problems and randomness; and 3) the optimization of doubly stochastic conditions is generally achieved by von Neumann successive projection (VNSP) lemma, which separates the conditions to form two subproblems for alternative optimization, converging only to a feasible solution. To solve these problems, in this article, a novel structured doubly stochastic graph-based clustering model termed SDSGC is proposed, which learns a structured doubly stochastic graph from data to directly provide cluster indicators. For optimization, a simple but effective augmented Lagrangian multiplier (ALM)-based method is proposed, which optimizes all the doubly stochastic conditions simultaneously to obtain the optimal solution. Experiments on one toy dataset and eight ad hoc noised face datasets have demonstrated that the proposed SDSGC is more robust to noise. Furthermore, a quantitative comparison of ten benchmarks has verified our SDSGC achieves better clustering performance when compared with SOTA methods. The code is available at https://github.com/NianWang-HJJGCDX/SDSGC.git.

Conventional scale-based major depression disorder (MDD) diagnosis methods are subjective, so it is significant to propose an objective and accurate MDD diagnosis method to assist physicians in diagnosing MDD. This paper proposes an MDD diagnostic method based on electroencephalogram (EEG) feature fusion and improved feature selection. First, seven functional connectivity matrices are extracted and reassembled into a vector to obtain the fusion functional connectivity feature. Then, a feature selection method based on principal component analysis, K-means, and mutual information (PKM) is constructed to optimize the high-dimensional EEG features. Finally, seven classifiers are used for MDD diagnosis. The results show that the proposed method performs better than the existing methods in MDD diagnosis with accuracy, sensitivity, and specificity of 88.73%, 90.67%, and 86%, respectively. Phase lag index (PLI) and phase-locked value (PLV) features, alpha and delta bands contribute significantly to MDD diagnosis. Functional connectivity in the right hemisphere of the brain, particularly in the right temporal and central prefrontal regions with other brain regions, may be beneficial for MDD diagnosis. High-precision MDD diagnosis can be achieved using EEG from only four channel pairs. In summary, this study provides an objective and accurate method for MDD diagnosis.

Depression electroencephalograph (EEG) classification based on machine learning is helpful for the auxiliary diagnosis of major depression disorder (MDD). Multi-channel EEG has abundant spatial information because EEG electrodes are distributed in different brain regions. However, existing methods arrange EEG features as feature vectors, which destroys the spatial structure of the features and may affect the model's performance. To improve the accuracy of MDD classification, we propose a novel EEG classification method for depression based on the brain topographic map and multi-scale convolutional transformer network (MCTNet). First, the power spectral density (PSD) features are extracted from EEG, and the one-dimensional feature vectors are converted into high-dimensional brain topographic maps according to the location information of EEG channels. Then, a multi-scale convolution with three parallel branches is designed to convert the brain topographic map into a deep feature map representation. Finally, image segmentation (IS) and the transformer encoder (TE) are used to learn the local and global features of the feature map, and the feature is input into the fully connected layer for classification. In addition, a joint loss function based on cross-entropy and center loss (CL) is designed to enable MCTNet to extract features with larger inter-class and smaller intra-class distances. Complete experimental verification is carried out on an open dataset. The accuracy, sensitivity and specificity of MCTNet are 97.24%, 97.20%, and 97.46%, respectively. The results show that the proposed method can achieve high-precision depression EEG classification and is superior to the state-of-the-art models.

Background: The single-channel sleep EEG has the advantages of convenient collection, high-cost performance, and easy daily use, and it has been widely used in the classification of sleep stages. Methods: This paper proposes a single-channel sleep EEG classification method based on long short-term memory and a hidden Markov model (LSTM-HMM). First, the single-channel EEG is decomposed using wavelet transform (WT), and multi-domain features are extracted from the component signals to characterize the EEG characteristics fully. Considering the temporal nature of sleep stage changes, this paper uses a multi-step time series as the input for the model. After that, the multi-step time series features are input into the LSTM. Finally, the HMM improves the classification results, and the final prediction results are obtained. Results: A complete experiment was conducted on the Sleep-EDFx dataset. The results show that the proposed method can extract deep information from EEG and make full use of the sleep stage transition rule. The proposed method shows the best performance in single-channel sleep EEG classification; the accuracy, macro average F1 score, and kappa are 82.71%, 0.75, and 0.76, respectively. Conclusions: The proposed method can realize single-channel sleep EEG classification and provide a reference for other EEG classifications.

With the rapid development of rule engine-based smart building systems, it is urgent to address the potential conflicts between different control rules to ensure system usability and reliability. For this end, this paper proposed a rule conflict detection approach based on the statistical model checking technique. Firstly, it presented a conflict taxonomy for smart building rules, systematically considering conflicts between different rules, and rule conflicts against system invariant constraints. Then, it proposed depicting smart building control rules, dynamic trigger sources and actuators with the NPTA (Network of Priced Timed Automata for short) model, and specifying rule conflicts with temporal logic. Finally, it proposed carrying out conflict detection with the statistical model checking technique based on the above formal model and formal specification. To validate the effectiveness of this approach, we carried out a case study on a superstore ventilation and air conditioning system, and demonstrated the rule conflict detection processes of 8 types of rule conflicts. The results demonstrate that the proposed approach can explicitly depict smart building rule interactions, and can effectively diagnose and detect rule conflicts, ensuring usability and reliability of the rule-based smart building control rules.

Brain network is often used to diagnose major depression disorder (MDD). However, existing methods require the artificial design of EEG features, which affects the recognition accuracy. This paper proposed a novel framework for MDD detection based on EEG functional connectivity and two-directional two-dimensional principal component analysis (TD-2DPCA). First, PLI was used to construct the functional connectivity matrix, which was then converted into a binary matrix and brain topology graph. On this basis, TD-2DPCA was used to extract features from the binary matrix, and the EEG features were input into different classifiers for MDD classification. Finally, experimental verification was carried out. The MDD detection accuracy of the proposed method was 96.55%, higher than that of graph theory, which verified the superiority of the proposed method.