Search

Hematologic diseases and blood disorders can be studied through microscopic examination of blood smear images or chemical assays. Many researchers are focused on utilizing deep learning (DL) to identify, quantify, and classify various types of blood cells, which is crucial for addressing theoretical and practical challenges in disease diagnosis and treatment planning. However, there is a significant deficiency in annotations for leukocyte classification, limiting the effectiveness of current DL methods. Contrastive learning (CL) facilitates the extraction of prior knowledge from unlabeled data, reducing the dependency on manual annotations. While CL demonstrates remarkable achievements in classifying natural images, the direct application to leukocyte classification presents certain defects. We observe that standard data augmentation techniques in CL exhibit varying sensitivity to different categories of white blood cell images. Employing a uniform augmentation strategy may lead the network into an optimization trap, acquiring inadequate representations from erroneous sample relationships. To address this issue, we propose a Re lative-Relationship-Guided C ontrastive L earning R epresentation (ReCLR) framework for the leukocyte classification. ReCLR mines positive and negative samples based on relative distance knowledge. Specifically, we provide adversarial guidance for the positive sample, restricting the distance between the positive sample and the original sample less than but as close as possible to the distance between the original sample and the farthest negative sample. Then, we utilize the entropy constraint to regulate the relative distance relationship between the negative and original samples. Finally, the guided positive and negative samples are employed for contrastive learning in leukocyte classification. The extensive experiments, including linear evaluation, domain transfer, and fine-tuning, demonstrate the effectiveness of the proposed method. Our ReCLR achieves accuracies of 92.07%, 65.36%, and 92.49% on three real-world leukocyte datasets, respectively, outperforming several state-of-the-art methods. The source code is released at https://github.com/AlchemyEmperor/ReCLR. [ABSTRACT FROM AUTHOR]

With the rapid development of autonomous driving technology, lane detection, a key component of intelligent vehicle systems, is crucial for ensuring road safety and efficient vehicle navigation. In this paper, a new lane detection method is proposed to address the problem of degraded performance of existing lane detection methods when dealing with complex road environments. The proposed method evolves from the original Cross Layer Refinement Network (CLRNet) by incorporating two of our carefully designed core components: the Global Feature Optimizer (GFO) and the Adaptive Lane Geometry Aggregator (ALGA). The GFO is a multi-scale attention mechanism that mimics the human visual focusing ability, effectively filtering out unimportant information and focusing on the image regions most relevant to the task. The ALGA is a shape feature-aware aggregation module that utilizes the shape prior of lanes to enhance the correlation of anchor points in an image, better fusing global and local information. By integrating both components into CLRNet, an enhanced version called Enhanced CLRNet (E-CLRNet) is presented, which exhibits higher performance stability in complex roadway scenarios. Experiments on the CULane dataset reveal that E-CLRNet demonstrates superior performance stability over the original CLRNet in complex scenarios, including curves, shadows, missing lines, and dazzling light conditions. In particular, in the curves, the F1 score of E-CLRNet is improved by almost 3% over the original CLRNet. This study not only improves the accuracy and performance stability of lane detection but also provides a new solution for the application of autonomous driving technology in complex environments, which promotes the development of intelligent vehicle systems. [ABSTRACT FROM AUTHOR]

Anomaly detection in time series data is crucial for many fields such as healthcare, meteorology, and industrial fault detection. However, traditional unsupervised time series anomaly detection methods suffer from biased anomaly measurement under contaminated training data. Most of existing methods employ hard strategies for contamination calibration by assigning pseudo-label to training data. These hard strategies rely on threshold selection and result in suboptimal performance. To address this problem, in this paper, we propose a novel unsupervised anomaly detection framework for contaminated time series (NegCo), which builds an effective soft contamination calibration strategy by exploiting the observed negative correlation between semantic representation and anomaly detection inherent within the autoencoder framework. We innovatively redefine anomaly detection in data contamination scenarios as an optimization problem rooted in this negative correlation. To model this negative correlation, we introduce a dual construct: morphological similarity captures semantic distinctions relevant to normality, while reconstruction consistency quantifies deviations indicative of anomalies. Firstly, the morphological similarity is effectively measured based on the representative normal samples generated from the center of the learned Gaussian distribution. Then, an anomaly measurement calibration loss function is designed based on negative correlation between morphological similarity and reconstruction consistency, to calibrate the biased anomaly measurement caused by contaminated samples. Extensive experiments on various time series datasets show that the proposed NegCo outperforms state-of-the-art baselines, achieving an improvement of 6.2% to 26.8% in A rea U nder the R eceiver O perating C haracteristics (AUROC) scores, particularly in scenarios with heavily contaminated training data. • Unsupervised time series anomaly detection under data contamination. • Calibrating the biased anomaly measurement by exploiting the negative correlation. • Normal samples from a learned Gaussian distribution to model negative correlation. • Single forward propagation enables anomaly detection using the trained autoencoder. [ABSTRACT FROM AUTHOR]

Deep learning techniques have greatly enhanced the performance of fire detection in videos. However, video-based fire detection models heavily rely on labeled data, and the process of data labeling is particularly costly and time-consuming, especially when dealing with videos. Considering the limited quantity of labeled video data, we propose a semi-supervised fire detection model called FireMatch, which is based on consistency regularization and adversarial distribution alignment. Specifically, we first combine consistency regularization with pseudo-label. For unlabeled data, we design video data augmentation to obtain corresponding weakly augmented and strongly augmented samples. The proposed model predicts weakly augmented samples and retains pseudo-label above a threshold, while training on strongly augmented samples to predict these pseudo-labels for learning more robust feature representations. Secondly, we generate video cross-set augmented samples by adversarial distribution alignment to expand the training data and alleviate the decline in classification performance caused by insufficient labeled data. Finally, we introduce a fairness loss to help the model produce diverse predictions for input samples, thereby addressing the issue of high confidence with the non-fire class in fire classification scenarios. The FireMatch achieved an accuracy of 76.92% and 91.80% on two real-world fire datasets, respectively. The experimental results demonstrate that the proposed method outperforms the current state-of-the-art semi-supervised classification methods. • Enhanced fire video classification via consistency regularization. • Mitigated sampling mismatch via adversarial distribution alignment. • Extensive comparison experiments with state-of-the-art methods. [ABSTRACT FROM AUTHOR]

• A classification method was developed for syndromes in traditional Chinese medicine. • Symptoms are informative clues to the identification of different syndromes. • Graph convolutional network can provide enhanced features for symptoms. • Besides symptoms, state elements can contribute to TCM syndrome classification. Traditional Chinese Medicine (TCM) diagnosis is a reasoning process through expert knowledge, in which syndrome classification is a key step for prescription recommendation and the treatment of patients. Doctors generally differentiate syndrome types according to patients' symptoms and state elements. This paper proposes a syndrome classification method based on graph convolutional network with residual structure, to exploit the potential relationship between symptoms and state elements. We constructed a graph convolutional network by combining symptoms and state elements for syndrome classification, called Symptoms-State elements Graph Convolutional Network (SSGCN), embedding the inherent logic of TCM diagnosis and treatment with a prescription graph. This graph architecture wherein contained the relationship between symptoms and state elements, and a multi-layer perceptron (MLP) was trained to classify different syndromes. Experiments were conducted on two self-built datasets according to two classic TCM books, i.e., Theories on Febrile Diseases and Traditional Chinese Medicine Prescription Dictionary. Accuracy, precision, recall and F1-score were adopted to evaluate the syndrome classificaiton results. Our proposed SSGCN method achieved accuracy of 75.59%, 69.63%, precision of 69.10%, 76.33%, recall of 75.63%, 66.67% and F1-score of 71.26%, 65.84% in the above two datasets, respectively. The proposed method for syndrome classification outperformed several popular methods including support vector machine, random forest, extreme gradient boosting and convolutional neural network. By constructing a prescription graph in which symptoms are used as nodes and state elements are taken into account for edges, graph convolution is implemted to capture the relationship of symptoms and state elements. This model improves the performance of syndrome classification and can be further extened for some other related applications in TCM. [ABSTRACT FROM AUTHOR]

Purpose To extract quantitative perfusion and texture features with computer assistance from contrast-enhanced ultrasound (CEUS) videos of breast cancer before and after neoadjuvant chemotherapy (NAC), and to evaluate pathologic response to NAC with these features. Methods Forty-two CEUS videos with 140,484 images were acquired from 21 breast cancer patients pre- and post-NAC. Time-intensity curve (TIC) features were calculated including the difference between area under TIC within a tumor and that within a computer-detected reference region (AUT_T-R). Four texture features were extracted including Homogeneity and Contrast. All patients were identified as pathologic responders by Miller and Payne criteria. The features between pre- and post-treatment in these responders were statistically compared, and the discrimination between pre- and post-treatment cancers was assessed with a receiver operating characteristic (ROC) curve. Results Compared with the pre-treatment cancers, the post-treatment cancers had significantly lower Homogeneity ( p < 0.001) and AUT_T-R ( p = 0.014), as well as higher Contrast ( p < 0.001), indicating the intratumoral contrast enhancement decreased and became more heterogeneous after NAC in responders. The combination of Homogeneity and AUT_T-R achieved an accuracy of 90.5% and area under ROC curve of 0.946 for discrimination between pre- and post-chemotherapy cancers without cross validation. The accuracy still reached as high as 85.7% under leave-one-out cross validation. Conclusions The computer-extracted CEUS features show reduced and more heterogeneous neovascularization of cancer after NAC. The features achieve high accuracy for discriminating between pre- and post-chemotherapy cancers in responders and thus are potentially valuable for tumor response evaluation in clinical practice. [ABSTRACT FROM AUTHOR]

Existing transition region-based image thresholding methods are unstable, and fail to achieve satisfactory segmentation accuracy on images with overlapping gray levels between object and background. This is because they only take the gray level mean of pixels in transition regions as the segmentation threshold of the whole image. To alleviate this issue, we proposed a robust hybrid single-object image segmentation method by exploiting salient transition region. Specifically, the proposed method first uses local complexity and local variance to identify transition regions of an image. Secondly, the transition region with the largest pixel number is chosen as salient transition region. Thirdly, a gray level interval is determined by using transition regions and image information, and one gray level of the interval is determined as the segmentation threshold by using the salient transition region. Finally, the image thresholding result is refined as final segmentation result by using the salient transition region to remove fake object regions. The proposed method has been extensively evaluated by experiments on 170 single-object real world images. Experimental results show that the proposed method achieves better segmentation accuracy and robustness than several types of image segmentation techniques, and enjoys its nature of simplicity and efficiency. [ABSTRACT FROM AUTHOR]

To search the global optimum across the entire search space with a very fast convergence speed, we propose a multi-strategy adaptive particle swarm optimization (MAPSO). MAPSO develops an innovative strategy of diversity-measurement to evaluate the population distribution, and performs a real-time alternating strategy to determine one of two predefined evolutionary states, exploration and exploitation, in each iteration. During iterative optimization, MAPSO can dynamically control the inertia weight according to the diversity of particles. Moreover, MAPSO introduces an elitist learning strategy to enhance population diversity and to prevent the population from possibly falling into local optimal solutions. The elitist learning strategy not only acts on the globally best particle, but also on some special particles that are very near to the globally best particle. The aforementioned features of MAPSO have been comprehensively analyzed and tested on eight benchmark problems and a standard test image. Experimental results show that MAPSO can substantially enhance the ability of PSOs to jump out of the local optimal solutions and significantly improve the search efficiency and convergence speed. [ABSTRACT FROM AUTHOR]

Existing transition region-based image thresholding is unsuitable for images with overlapping gray levels between object and background due to the essence of global thresholding. To alleviate this issue, we proposed an innovative transition region-based single-object image segmentation method. The proposed algorithm first extracted transition regions of an image by using local variance as the descriptor. Its second step, image thinning, was to skeletonize transition regions as single pixel edges. The third step, edge filtering, removed useless short edges and edge spikes. Subsequently, the step called edge linking connected interrupted edges to obtain closed object contours. The final step filled object regions confined by the object contours with black or white, and only the largest object region remained as the final image segmentation result. The proposed algorithm was compared with different types of image segmentation methods on a variety of real world images, and experimental results demonstrated its superiority. [ABSTRACT FROM AUTHOR]

• An expert-knowledge attention network (EKANet) was designed to classify four tachyarrhythmia on electrocardiogram (ECG) signals. • Two attention modules in EKANet are explored for abnormal rhythm representation and multiple leads selection. They are guided by the knowledge of cardiology clinicians. • The two attention modules in EKANet required no extra training process. The low complex structure increases the generalization and robustness of EKANet. • EKANet can be more generalized and robust to different ECG datasets than the comparison methods. An expert-knowledge attention network (EKANet) was designed to improve the accuracy of arrhythmia diagnosis and reduce the recheck time. This network classifies four tachyarrhythmia on electrocardiogram (ECG) signals, encompassing most arrhythmia diseases. In the EKANet, two attention modules based on the knowledge of cardiology can rapidly capture the ECG rhythm and P waves in multiple leads without any training. This mechanism is performed to reduce the computational time of re-building a model. The EKANet integrates a six-layer convolutional neural network (CNN) and a gated recurrent unit (GRU) as the classifier to realise the tachyarrhythmia classification. The EKANet outperformed 1D CNN and ArrhythmiaNet on the MIT-BIH datasets by 3.1% on average accuracy. Furthermore, the EKANet achieved approximately 8.5% and 3.9% average F 1-score increases on the dataset of China ECG challenge contest compared with time-incremental CNN (TI-CNN) and attention-based TI-CNN, respectively. Meanwhile, the EKANet has a much lower complexity than that of the other typical models with a competitive accuracy. [ABSTRACT FROM AUTHOR]

Abstract: Classic statistical thresholding methods based on maximizing between-class variance and minimizing class variance fail to achieve satisfactory results when segmenting a kind of image, where variance discrepancy between the object and background classes is large. The reason is that they take only class variance sum of some form as criterions for threshold selection, but neglect discrepancy of the variances. In this paper, a novel criterion combining the above two factors is proposed to eliminate the described limitation for classic statistical approaches and improve segmentation performance. The proposed method determines the optimal threshold by minimizing the criterion. The method was compared with several classic thresholding methods on a variety of images including some NDT images and laser cladding images, and the experimental results show the effectiveness of the algorithm. [ABSTRACT FROM AUTHOR]

Abstract: Thresholding method based on transition region is a newly developed approach for image segmentation in recent years. In this paper, a novel transition region extraction and thresholding method based on gray level difference is proposed by analyzing properties of transition region. The gray level difference can effectively represent the essence of transition region. Hence, the proposed algorithm can accurately extract transition region of an image and get ideal segmentation result. The proposed algorithm was compared with two classic transition region-based methods on a variety of synthetic and real world images, and the experimental results show the effectiveness and efficiency of the algorithm. [Copyright &y& Elsevier]

Abstract: Different types of cloud amounts have different influences on the calculation of the radiation energy of the soil–air system. Based on the occurrence of the overlaying of high, medium and low types of clouds and the interaction and interrelationship of information provided by them, this paper puts forwards new concepts of “effective” frequency of occurrences of cloud and “effective” time-average cloud amount and their calculation formulas, while the commonly used calculation formula for time-average cloud amount is treated as a special case of the calculation formula for “effective” time-average cloud amount. Through a principal component analysis of the correlation coefficient matrix formed by conditional probabilities exhibited by any other type of cloud when a certain type of cloud occurs, the normalized weight of the conditional probability exhibited by any other type of cloud when this particular type of cloud occurs is ascertained. When this particular type of cloud is obscured by another type of cloud, the weight can be seen as the probability of this particular type of cloud being obscured by another type of cloud. With the formula for the “effective” frequency of occurrences of cloud and the formula for “effective” average cloud, this probability can be used to calculate the “effective” time-average cloud amount of this particular type of cloud. By applying this method to the analysis and calculation of the “effective” average cloud amount in the four different seasons as observed by 15 observation stations worldwide, this paper finds that the sum of the “effective” time-average cloud amounts of each type of clouds is far closer to the observed value of the total cloud amount than the sum of the time-average cloud amounts derived by using only the calculation formula for time-average cloud amount, with more than 87% of the relative errors of the total cloud amount corrections made by the former being smaller than 15% and all relative errors of the total average cloud corrections made by the latter exceeding 15%. [Copyright &y& Elsevier]

• Challenge of WRCC forecasting is the characteristics of nonlinear multiple indicators. • A feedforward neural network based on normalization and error correction is proposed. • Units and values of multiple indicators are normalized simultaneously. • An error correction method is adopted to improve the effectiveness of proposed model. • Final calibrated model is expressed as an equation rather than a programming language. The water resources carrying capacity (WRCC) is fundamental in aiding sustainable socioeconomic regional development, and increasing attention is being paid to WRCC forecasting. The challenge of WRCC forecasting lies in the complex characteristics of nonlinear multiple indicator time series data. To address this problem, this study proposed a feedforward neural network (FNN) based on normalization and error correction for WRCC forecasting. Firstly, units and values of multiple indicators for use in the WRCC were normalized simultaneously to allow the data to be treated as a single equivalent indicator. Thus, the high-dimensional forecasting model was simplified to a low-dimensional model. To improve the effectiveness of the model, an error correction method was then adopted to correct forecasted WRCC values according to similar corresponding sample values. Two simple types of structured FNN were used to address the overfitting problem, and the bipolar sigmoid function was used as the activation function of hidden nodes. The final calibrated model was expressed as an equation rather than a programming language, thus making it easier to use. Yantai, a city in Shandong Province, was selected as a case study to validate this proposed method. Results showed that the mean relative absolute errors of these two simplified FNN models are 1.23% and 1.18%, respectively. Compared to other models, this model has been shown to be feasible and simple to use for WRCC forecasting. [ABSTRACT FROM AUTHOR]

Syndrome classification plays a key role in the clinical diagnosis and treatment with traditional Chinese medicine (TCM), aiming to identify the disease type. Given symptoms of a patient, existing approaches for syndrome classification are generally limited to modeling the interaction between symptoms and syndromes while ignoring the induction of state elements. To alleviate this issue, a state-element-aware hypergraph convolutional network (SEHGCN) is proposed to incorporate state elements into syndrome classification and discover high-order semantic relationships among TCM entities through hypergraph convolutional network (HGCN). Specifically, state elements are initially induced from symptoms by an extraction network, then symptoms and state elements are embedded via convolution on patient hypergraph to obtain the latent representation of patients. Finally, syndromes are classified by a multilayer perceptron (MLP). Extensive experiments on two TCM datasets show that the syndrome classification results with this proposed method are significantly improved over other competing methods. [ABSTRACT FROM AUTHOR]

• An unsupervised prompt learning method is proposed to identify the target object. • Our method can capture important feature regions through four dimensions. • It can improve the diversity and discriminability of representation. • A general hybrid PCA-whitening (HPW) method is proposed to improve performance. Identifying the target object in an image can produce more accurate and discriminative feature representations, which can significantly improve large-scale instance-level image-retrieval performance. However, it is usually difficult to obtain annotation information for all target objects in a dataset manually, which makes it challenging to automatically identify target objects. To address this issue, we propose a novel method of instance-level image retrieval based on unsupervised prompt learning and regional attention (PLRA) rather than manual annotation. It includes three main components: (1) We propose an unsupervised prompt learning method to identify an image's target object. It reconstructs deep features by mining prompt information, then designs prompt factors to identify the target object based on the reconstructed features. (2) We propose a new regional attention method to extract the distinguishing features of the target object. This method captures important feature regions in four dimensions: global, local, spatial, and channel, which improves the diversity and discriminability of the representation. (3) We propose a general hybrid PCA-whitening (HPW) method based on multi-parameter learning and feature fusion, which trades off feature dimensionality with retrieval performance. This method significantly improves performance and reduces vector dimensionality in a plug-and-play manner. Comprehensive experiments on five benchmark datasets show that the proposed method significantly outperforms existing state-of-the-art methods based on unsupervised feature aggregation. [ABSTRACT FROM AUTHOR]

Image dedusting has gained increasing attention as a preprocessing step for computer vision tasks. Current traditional image dedusting methods rely on a variety of constraints or priors, which are easy to be limited in real complex dusty scenes. Recently, some paired learning CNN methods have been proposed. However, they usually imitate the idea of image dehazing and fail to consider the inherent natural characteristics of dusty images. To bridge this gap, we propose an unpaired learning algorithm for image dedusting based on Retinex with GANs (DedustGAN). Specifically, we first design the DedustGAN with two branches, one branch uses Retinex to obtain one preliminary dedusted image in a physical way, and the other branch uses GANs to obtain another preliminary dedusted image in a learning way. Then, we design a no-reference image fusion strategy to fuse the two preliminary dedusted images for outputting the final dedusted image. The proposed DedustGAN breaks the restriction of existing paired learning methods that require paired images for network training and provide an effective physical dedusting model. Qualitative and quantitative experimental comparisons demonstrate the powerful dedusting ability of the Retinex model and the superiority of DedustGAN over the existing methods. In addition, to validate the strong generalization of the proposed DedustGAN, we further performed experiments on underwater image enhancement and low-light image enhancement tasks. We found that the effectiveness of DedustGAN even outperforms the existing methods in their domains. The code of the proposed DedustGAN is available at https://github.com/MXL696/DedustGAN. • DeustGAN based on Retinex with GANs can train with unpaired clear-dusty images. • we proved the significant dust removal performance of the Retinex. • DedustGAN designed a no-reference fusion strategy for various enhancement tasks. • Experimental results show that DedustGAN provides better dedusting performance. [ABSTRACT FROM AUTHOR]

Deep convolutional neural networks (CNNs) have attracted considerable interest in low-level computer vision. Researches are usually devoted to improving the performance via very deep CNNs. However, as the depth increases, influences of the shallow layers on deep layers are weakened. Inspired by the fact, we propose an attention-guided denoising convolutional neural network (ADNet), mainly including a sparse block (SB), a feature enhancement block (FEB), an attention block (AB) and a reconstruction block (RB) for image denoising. Specifically, the SB makes a tradeoff between performance and efficiency by using dilated and common convolutions to remove the noise. The FEB integrates global and local features information via a long path to enhance the expressive ability of the denoising model. The AB is used to finely extract the noise information hidden in the complex background, which is very effective for complex noisy images, especially real noisy images and bind denoising. Also, the FEB is integrated with the AB to improve the efficiency and reduce the complexity for training a denoising model. Finally, a RB aims to construct the clean image through the obtained noise mapping and the given noisy image. Additionally, comprehensive experiments show that the proposed ADNet performs very well in three tasks (i.e. synthetic and real noisy images, and blind denoising) in terms of both quantitative and qualitative evaluations. The code of ADNet is accessible at https://github.com/hellloxiaotian/ADNet. [ABSTRACT FROM AUTHOR]

• A hybrid binary particle swarm optimization is proposed. • Applying an adaptive penalty function as fitness function. • Using a tabu search procedure to improve solution quality. • Presenting a tabu based mutation procedure to achieve diversification. • We find new best solutions for 28 out of 30 instances. The set-union knapsack problem (SUKP) is a generalization of the standard 0–1 knapsack problem. It is NP-hard, and has several industrial applications. Several approximation and heuristic approaches have been previously presented for solving the SUKP. However, the solution quality still needs to be enhanced. This work develops a hybrid binary particle swarm optimization with tabu search (HBPSO/TS) to solve the SUKP. First, an adaptive penalty function is utilized to evaluate the quality of solutions during the search. This method endeavours to explore the boundary of the feasible solution space. Next, based on the characteristics of the SUKP, a novel position updating procedure is designed. The newly generated solutions obtain the good structures of previously found solutions. Then, a tabu based mutation procedure is introduced to lead the search to enter into new hopeful regions. Finally, we design a tabu search procedure to improve the exploitation ability. Furthermore, a gain updating strategy is employed to reduce the solution time. The HBPSO/TS is tested on three sets of 30 benchmark instances, and comparisons with current state-of-the-art algorithms are performed. Experimental results show that HBPSO/TS performs much better than the other algorithms in terms of solution quality. Moreover, HBPSO/TS improves new best results at 28 out of the 30 instances. The impact of the main parts of the HBPSO/TS is also experimentally investigated. [ABSTRACT FROM AUTHOR]

Least square regression is a very popular supervised classification method. However, two main issues greatly limit its performance. The first one is that it only focuses on fitting the input features to the corresponding output labels while ignoring the correlations among samples. The second one is that the used label matrix, i.e. , zero–one label matrix is inappropriate for classification. To solve these problems and improve the performance, this paper presents a novel method, i.e. , inter-class sparsity based discriminative least square regression (ICS_DLSR), for multi-class classification. Different from other methods, the proposed method pursues that the transformed samples have a common sparsity structure in each class. For this goal, an inter-class sparsity constraint is introduced to the least square regression model such that the margins of samples from the same class can be greatly reduced while those of samples from different classes can be enlarged. In addition, an error term with row-sparsity constraint is introduced to relax the strict zero–one label matrix, which allows the method to be more flexible in learning the discriminative transformation matrix. These factors encourage the method to learn a more compact and discriminative transformation for regression and thus has the potential to perform better than other methods. Extensive experimental results show that the proposed method achieves the best performance in comparison with other methods for multi-class classification. [ABSTRACT FROM AUTHOR]

It is a very challenging problem to well simulate visual attention mechanisms for content-based image retrieval. In this paper, we propose a novel computational visual attention model, namely saliency structure model, for content-based image retrieval. First, a novel visual cue, namely color volume, with edge information together is introduced to detect saliency regions instead of using the primary visual features (e.g., color, intensity and orientation). Second, the energy feature of the gray-level co-occurrence matrices is used for globally suppressing maps, instead of the local maxima normalization operator in Itti׳s model. Third, a novel image representation method, namely saliency structure histogram, is proposed to stimulate orientation-selective mechanism for image representation within CBIR framework. We have evaluated the performances of the proposed algorithm on two datasets. The experimental results clearly demonstrate that the proposed algorithm significantly outperforms the standard BOW baseline and micro-structure descriptor. [ABSTRACT FROM AUTHOR]

The recommendation of herbal prescriptions is a focus of research in traditional Chinese medicine (TCM). Artificial intelligence (AI) algorithms can generate prescriptions by analysing symptom data. Current models mainly focus on the binary relationships between a group of symptoms and a group of TCM herbs. A smaller number of existing models focus on the ternary relationships between TCM symptoms, syndrome-types and herbs. However, the process of TCM diagnosis (symptom analysis) and treatment (prescription) is, in essence, a "multi-ary" (n -ary) relationship. Present models fall short of considering the n -ary relationships between symptoms, state-elements, syndrome-types and herbs. Therefore, there is room for improvement in TCM herbal prescription recommendation models. To portray the n -ary relationship, this study proposes a prescription recommendation model based on a multigraph convolutional network (MGCN). It introduces two essential components of the TCM diagnosis process: state-elements and syndrome-types. The MGCN consists of two modules: a TCM feature-aggregation module and a herbal medicine prediction module. The TCM feature-aggregation module simulates the n -ary relationships between symptoms and prescriptions by constructing a symptom-'state element'-symptom graph (S e) and a symptom-'syndrome-type'-symptom graph (T s). The herbal medicine prediction module inputs state-elements, syndrome-types and symptom data and uses a multilayer perceptron (MLP) to predict a corresponding herbal prescription. To verify the effectiveness of the proposed model, numerous quantitative and qualitative experiments were conducted on the Treatise on Febrile Diseases dataset. In the experiments, the MGCN outperformed three other algorithms used for comparison. In addition, the experimental data shows that, of these three algorithms, the SVM performed best. The MGCN was 4.51%, 6.45% and 5.31% higher in Precision@5, Recall@5 and F1-score@5, respectively, than the SVM. We set the K -value to 5 and conducted two qualitative experiments. In the first case, all five herbs in the label were correctly predicted by the MGCN. In the second case, four of the five herbs were correctly predicted. Compared with existing AI algorithms, the MGCN significantly improved the accuracy of TCM herbal prescription recommendations. In addition, the MGCN provides a more accurate TCM prescription herbal recommendation scheme, giving it great practical application value. [Display omitted] [ABSTRACT FROM AUTHOR]

Many early ophthalmology-related diseases can be detected through fundus images. The cup-to-disc ratio (CDR) is an important criterion for the early screening and diagnosis of glaucoma. However, accurately calculating it is a challenging task primarily due to blurred boundaries and the physiological structure of the fundus. To alleviate these issues, this paper designs a location-constrained joint optic disc (OD) and optic cup (OC) segmentation method using the multiplex aggregation network (LC-MANet). This method incorporates independent segmentation and joint segmentation within a coarse-to-fine framework. In the independent segmentation stage, this paper utilizes the initial OD and OC as prior location auto-context information to constrain the joint segmentation of OD and OC. The proposed multiplex aggregation network combines features from the corresponding layer and across layers to capture global and local information. This paper employs a multi-channel fusion strategy in image preprocessing to suppress vascular interference in OD independent segmentation. To emphasize the superiority of our method, it is important to highlight the significant number of test images used in the evaluation. Our experiments are conducted on three public datasets, namely RIM-ONE (60 images), Drishti-GS (51 images), and REFUGE (400 images). Experimental results demonstrate that our method outperforms most others in terms of Dice Coefficient and absolute error. The accuracy of glaucoma screening achieved by our proposed method for the Drishti-GS, REFUGE, and RIM-ONE-r3 datasets is 0.9216, 0.9600, and 0.9000, respectively. The proposed framework successfully designs a multiplex aggregation network for a coarse-to-fine joint segmentation model and achieves state-of-the-art performance in OD and OC segmentation. Furthermore, it can be applied to other medical image segmentation tasks. [ABSTRACT FROM AUTHOR]

Occluded person re-identification (Re-ID) is a challenging task, as pedestrians are often obstructed by various occlusions, such as non-pedestrian objects or non-target pedestrians. Previous methods have heavily relied on auxiliary models to obtain information in unoccluded regions, such as human pose estimation. However, these auxiliary models fall short in accounting for pedestrian occlusions, thereby leading to potential misrepresentations. In addition, some previous works learned feature representations from single images, ignoring the potential relations among samples. To address these issues, this paper introduces a Multi-Level Relation-Aware Transformer (MLRAT) model for occluded person Re-ID. This model mainly encompasses two novel modules: Patch-Level Relation-Aware (PLRA) and Sample-Level Relation-Aware (SLRA). PLRA learns fine-grained local features by modeling the structural relations between key patches, bypassing the dependency on auxiliary models. It adopts a model-free method to select key patches that have high semantic correlation with the final pedestrian representation. In particular, to alleviate the interference of occlusion, PLRA captures the structural relations among key patches via a two-layer Graph Convolution Network (GCN), effectively guiding the local feature fusion and learning. SLRA is designed to facilitate the model to learn discriminative features by modeling the relations among samples. Specifically, to mitigate noisy relations of irrelevant samples, we present a Relation-Aware Transformer (RAT) block to capture the relations among neighbors. Furthermore, to bridge the gap between training and testing phases, a self-distillation method is employed to transfer the sample-level relations captured by SLRA to the backbone. Extensive experiments are conducted on four occluded datasets, two partial datasets and two holistic datasets. The results show that the proposed MLRAT model significantly outperforms existing baselines on four occluded datasets, while maintains top performance on two partial datasets and two holistic datasets. • A MLRAT model is introduced to learn robust feature representations. • A PLRA module is designed to identify key patches and perform local feature learning. • An SLRA module is presented to learn discriminative features. [ABSTRACT FROM AUTHOR]

The detection of cracks is essential for assessing and maintaining building and road safety. However, the large appearance variations and the complex topological structures of cracks bring challenges to automatic crack detection. To alleviate the above challenges, we propose a deep multi-scale crack feature learning model called DeepCrackAT for crack segmentation, which is based on an encoder–decoder network with feature tokenization mechanism and attention mechanism. Specifically, we use hybrid dilated convolutions in the first three layers of the encoder–decoder to increase the network's receptive field and capture more crack information. Then, we introduce a tokenized multilayer perceptron (Tok-MLP) in the last two layers of the encoder–decoder to tokenize and project high-dimensional crack features into low-dimensional space. This helps to reduce parameters and enhance the network's ability of noise resistance. Next, we concatenate the features corresponding to the encoder–decoder layers and introduce the convolutional block attention module (CBAM) to enhance the network's perception of the critical crack region. Finally, the five-layer features are fused to generate a binary segmentation map of the crack image. We conducted extensive experiments and ablation studies on two real-world crack datasets, and DeepCrackAT achieved 97.41% and 97.25% accuracy on these datasets, respectively. The experimental results show that the proposed method outperforms the current state-of-the-art methods. [ABSTRACT FROM AUTHOR]

• We propose a strategy to adaptively estimate minimal structure sizes by using previously obtained minimal structure sizes. • We propose another strategy to generate effective initial model hypotheses by jointly performing feature matching and proximity sampling. • We present an efficient sampling algorithm based on the above two proposed strategies. • Extensive experimental results demonstrate the effectiveness of the proposed sampling algorithm. Greedy search-based guided sampling is a promising research field in model fitting to data with multiple structures in the presence of a large number of outliers. However, these greedy search-based guided sampling algorithms are sensitive to the fixed minimal (acceptable) structure size and the initial model hypothesis: when the fixed minimal structure size is too small, data subsets sampled by these algorithms are not representative. In contrast, when it is too large, data subsets might be contaminated by outliers. Furthermore, these algorithms may fail to obtain an accurate model hypothesis, if the initial model hypothesis is far from the true model. In this paper, we address the above-mentioned two issues by proposing two greedy search-based strategies: one aims to adaptively estimate minimal structure sizes and the other aims to generate effective initial model hypotheses. Specifically, on one hand, to avoid using the fixed minimal structure size, a strategy is proposed to adaptively estimate minimal structure sizes by using previously obtained ones. On the other hand, to reduce the impact of outliers, a strategy is proposed to filter out outliers to obtain a reduced data subset by using a feature matching algorithm. Then, this strategy generates promising initial model hypotheses by using a proximity sampling on the reduced data subset. Finally, an efficient sampling algorithm based on the two proposed greedy search-based strategies is applied to three vision tasks, i.e., fundamental matrix estimation, homography plane detection and 3D motion segmentation. Extensive experimental results demonstrate the effectiveness of the proposed sampling algorithm. [ABSTRACT FROM AUTHOR]

Abstract: Numerous computational models of visual attention have been suggested during the last two decades. But, there are still some challenges such as which of early visual features should be extracted and how to combine these different features into a unique “saliency” map. According to these challenges, we proposed a sparse embedding visual attention system combined with edge information, which is described as a hierarchical model in this paper. In the first stage, we extract edge information besides color, intensity and orientation as early visual features. In the second stage, we present a novel sparse embedding feature combination strategy. Results on different scene images show that our model outperforms other visual attention computational models. [Copyright &y& Elsevier]