Millimeter-wave radar object detection has become pivotal for autonomous driving systems requiring all-weather reliability. While conventional CFAR methods face limitations in classification capability and require extensive preprocessing, existing deep learning approaches exhibit inherent constraints: CNNs struggle with global context modeling due to limited receptive fields, while Transformers demand substantial labeled data—a critical bottleneck given radar annotation costs. To address these dual challenges, we propose UniROD, a unified framework integrating MAE-based self-supervised learning with hierarchical CNN-Transformer fusion. Our architecture introduces two key innovations: (1) Adaptive Position Embedding (APE) that dynamically encodes spatial relationships in radar spectrograms, and (2) Hierarchical Mixing (HM) blocks combining CNN layers for local feature extraction with Swin Transformer layers for global dependency modeling. Employing a 75% window masking strategy during MAE pre-training, the model effectively learns transferable representations from unlabeled data. Comprehensive evaluations demonstrate UniROD’s state-of-the-art performance with 71.6% [email protected] and 58.1% [email protected] on RADDet, along with 84.83% AP and 88.25% AR on CRUW. These advancements demonstrate significant potential for label-efficient radar perception in real-world deployments.

The measurement of ocular alignment is critical for the diagnosis of strabismus. Current clinical methods for assessing ocular misalignment are subjective and frequently rely on the expertise of practitioners and the extent of patient cooperation. Computer-aided diagnosis methods in recent years have improved automation and precision of measurement, but still, fall short of the requirement of clinical practice. In this study, a video-based intelligent ocular misalignment assessment (VIOMA) system, was proposed to provide an objective, repeatable, user-friendly and highly-automated alternative modality for clinical ocular misalignment measurement, in which the automatic cover tests were performed under a control and motor unit, simultaneously the eye movements were tracked using a motion-capture module and assessed through video analysis techniques, determining the presence, type, and magnitude of eye deviation. For system evaluation, an automatic cover tests video dataset for strabismus (StrabismusACT-76) was established, which consists of data from 76 participants. The Bland-Altman plot, used to compare the results of the VIOMA system and human expert, showed a mean value of 1.26 prism diopter (PD) and a half-width of the 95% limit of agreement of ±7.17 PD. VIOMA system presented a mean absolute error of 3.04 PD in measuring the deviation magnitude, within a 5 PD error tolerance. Additionally, the system’s measurements were strongly correlated with that of video labeling with the mean value of -0.26 PD, a half-width of the 95% limit of agreement of ±3.56, and the average error of 1.31 PD. The experiment results indicated that the proposed method has the capability to offer accurate and efficient assessment of ocular misalignment. Note to Practitioners—The motivation behind this work stems from the need to develop an accurate and efficient system for automated ocular misalignment assessment. The subjectivity of manual cover test performed by examiners has led to variability in outcomes, and certain existing computerized methods have limitations in terms of automation, measurement accuracy, and applicability in clinical practice. Faced with these challenges, we proposed VIOMA, by establishing an apparatus for automatic implementation of cover tests and developing assessment algorithms based on strabismic video analysis. This system can objectively and precisely measure ocular misalignment, offering a promising practical solution for clinical intelligent diagnosis of strabismus. The VIOMA system’s potential applications are not limited to strabismus but may extend to other eye-related conditions and beyond.

Accurate coronary artery segmentation is crucial for quantitative analysis of coronary arteries in noninvasive coronary computed tomography angiography (CCTA) images. However, current segmentation algorithms often have unsatisfactory recall due to the small size and complex morphology of coronary arteries, particularly in the distal segments. To address this issue, we introduce a new fully automated method named Ensembled-SAMs, which harnesses the strengths of the Segment Anything Model (SAM) and the no-new-U-Net (nnU-Net). First, noisy bounding box prompts are automatically generated by a vesselness algorithm that highlights the tubular structures in the CCTA images. These noisy prompts are then used to fine-tune the SAM and its two variants separately. The SAM variants introduce a classification head in their mask decoder to alleviate the false positives. In addition, an nnU-Net segmentation network is trained from scratch. Finally, the outputs of the SAMs and the nnU-Net are strategically aggregated to obtain the final segmentation result. Experiments on both a self-built dataset and the public Automated Segmentation of Coronary Arteries (ASOCA) challenge dataset demonstrate that the proposed Ensembled-SAMs outperforms the state-of-the-arts, achieving precise segmentation of coronary arteries, with particular enhancement in delineating small coronary artery segments.

Accurate spatial localization and classification of targets are crucial for achieving target detection based on millimeter-wave automotive radar. However, the lower angular resolution hinders the perceptual capabilities of radar systems, making it challenging to meet the application requirements in real-world environments. This paper first quantitatively analyzes the impact of angular resolution on the performance of downstream target detection tasks on radar data from real road scenarios. Then a radar angular super-resolution algorithm based on Multiple Input Multiple Output (MIMO) radar virtual receive channel augmentation is proposed, without increasing the complexity and cost of the radar hardware system. Validation on downstream target detection tasks demonstrates that radar virtual receive channel augmentation effectively increases the radar system’s angular resolution, thereby enhancing the performance of downstream target detection tasks.

The intricate and dynamic nature of eye movements serves as a window into the realms of cognition, emotion, and physiological responses. Event detection, in turn, is instrumental in the precise recognition and categorization of these diverse eye movements. Deep learning methods have recently been applied to event detection, yielding promising results. However, the intrinsic multi-scale attributes of events have often been overlooked in existing approaches. To address this, we introduce “GazeUNet”, a novel network based on U-Net and Bi-GRU, which classifies gaze samples into three categories: fixation, saccade, and post-saccadic oscillations. Firstly, multi-scale spatial features are captured using a U-Net model, and then a hierarchical bidirectional gated recurrent unit (Bi-GRU) is employed to extract temporal correlations, followed by classification through fully connected layers. Our results, derived from the analysis of three publicly available datasets, consistently showcase the superiority of the proposed model compared with other state-of-the-art methods across all categories.

Coronary artery disease (CAD), a leading cause of mortality and morbidity, manifests as atherosclerotic plaques formed by the deposition of cholesterol and lipids within coronary walls. A plenty of machine learning methods have been developed to identify different plaques or the degree of stenosis. Few studies, however, focus on plaques vulnerability, which is crucial because vulnerable plaques are at a high risk of rupture or erosion even with less severe stenosis. To this end, we propose a weakly-supervised multi-contrastive learning network named WheelNet to differentiate vulnerable plaques from stable ones in coronary CT angiography (CCTA). Specifically, we first extract cross-sectional images along the coronary centerline and took consecutive cross-sections as one image sequence. Second, we construct a WheelNet with multiple branches to perform contrastive learning between different image sequences, dependent or independent of vulnerability labels of coronary plaques. Third, we perform patient-level feature aggregation via local-to-global feature encoding given the feature embeddings of image sequences. Finally, we differentiae patients with vulnerable coronary plaques from those with stable ones using an XGBoost classifier. Extensive experiments on the CCTA dataset of 108 patients show the superiority of our WheelNet over other state of the arts, with the diagnostic area-under-the-curve (AUC) of 0.74/0.75 with/without using vulnerability labels, respectively.

Neural architecture search (NAS) adopts a search strategy to explore the predefined search space to find superior architecture with the minimum searching costs. Bayesian optimization (BO) and evolutionary algorithms (EA) are two commonly used search strategies, but they suffer from being computationally expensive, challenging to implement, and exhibiting inefficient exploration ability. In this article, we propose a neural predictor guided EA to enhance the exploration ability of EA for NAS (NPENAS) and design two kinds of neural predictors. The first predictor is a BO acquisition function for which we design a graph-based uncertainty estimation network as the surrogate model. The second predictor is a graph-based neural network that directly predicts the performance of the input neural architecture. The NPENAS using the two neural predictors are denoted as NPENAS-BO and NPENAS-NP, respectively. In addition, we introduce a new random architecture sampling method to overcome the drawbacks of the existing sampling method. Experimental results on five NAS search spaces indicate that NPENAS-BO and NPENAS-NP outperform most existing NAS algorithms, with NPENAS-NP achieving state-of-the-art performance on four of the five search spaces.

Poleward moving auroral forms (PMAFs) are a common dayside auroral phenomenon, and the study of PMAFs has important implications for the exploration of the near-earth space physical processes for geosciences. In the all-sky imager (ASI) image sequence, PMAFs show a tendency to move northward in the northern hemisphere. Therefore, this particular motion pattern can be used for PMAF recognition. Previous works for automatic recognition of PMAFs tend to rely on optical flow. However, both the traditional and the deep learning-based optical flow estimation methods are time- and memory-expensive. In view of the large number of auroral images generated every year, it is impractical to estimate the optical flow for all auroral data with limited computational resources. In this letter, a poleward-motion aware network (PA-Net) is proposed to extract the motion features directly from ASI images. PA-Net computes the correlation between each point in an image and the points at the poleward direction in the following image by means of a poleward-motion aware operation (PA-Operation), to verify whether the point under consideration has undergone poleward motion. In addition, a channel attention mechanism is applied to the features obtained by PA-Operation to suppress information less helpful for recognizing PMAFs. The PA-Net achieves the best performance on the PMAFs recognition dataset over other commonly used action recognition models, validating the superiority of our approach. More importantly, the complicated optical flow estimation is avoided, making it possible to apply the proposed method to large-scale auroral data.

Neuron detection is a key step in individualizing and counting neurons which are important for assessing physiological and pathophysiological information. A large number of methods including deep learning networks have been proposed but mainly targeting regions with few aggregated neurons. The objective of this paper is to address an automated neuron detection problem in heterogeneous hippocampus region with different degrees of neuron aggregation. Since deep learning networks require a lot of ground truths but neuron instance annotation is impossible in regions where numerous neurons are clustered, ground truth of centroids marked at the center of neurons is created for training. We propose a multiscale convolutional neural network (CNN) to regress neuron centroid mapping across image. Using multiscale information makes the proposed network applicable not only for single individual neurons, but also for a large number of aggregated neurons. Experimental results show that our method is superior to state-of-the-art deep learning-based algorithms. To our knowledge, this is the first deep learning study to detect neurons in regions of highly clustered neurons.

The width of auroral arc is one of the most important factors in understanding and examining its physical mechanisms. In this letter, we propose a fully automatic method for computing the width of auroral arcs based on the instance segmentation of auroral images. To accurately detect and segment auroral arcs with oriented bounding boxes, we adapt a state-of-the-art instance segmentation model, Mask region-based convolutional neural network, by designing a two-stage inference process combined with an indispensable random rotation training strategy and designing an effective feature extraction architecture. Given the segmented masks of individual auroral arcs, we present a method for computing the arc width automatically. In our experiments, the instance segmentation model achieves 86.8% of mean average precision on the human-labeled data set. By automatically evaluating the width of 29 938 detected auroral arcs in 18 417 auroral arc images, we obtain a similar arc width distribution to that evaluated by the semiautomatic approach, which demonstrates the effectiveness of our proposed method.

Big data-driven technologies and deep learning approaches are being drawn much attention to Coronary Heart Disease(CHD) early identification and risk warning research. CHD is one of the common chronic diseases that threaten the health and life of people. Cohort study method and machine learning method are often used to identify to target the patients precisely. To the best of our knowledge, the literatures mostly focused on how to establish and optimize the identification and warning models or the cohort study, while overlooking the data management. To promote the early identification and risk warning research of CHD, we contribute a cooperated data management platform in regards to the big patient data and big CHD early identification model data. According to the characteristics of the model data, we propose the SMR(Samples-Model-Results) data chain conception to describe the relationship among the training data, model and the model evaluation result. The conceptual schema about CHD patient cohort and CHD early identification model are abstracted which are system-independent representations. To target the DBMS, system-dependent logical data schemas are designed based on the conceptual data model. The experiments about the efficiency of relational database and NoSQL database based solutions are conducted. To manage the CHD early identification model data effectively, we propose the model version to represent the relationship between the models considering the modeling lifecycle. The model tree is established and the query algorithms are designed to perform the lineage management of the CHD early identification models. The effective patient data visual exploration services, cohort study services and CHD early identification model selection, model comparison and model data visual exploration services are implemented for CHD early identification and risk warning researchers based on the architecture design of the Cooperated Data Management Platform.

Machine learning task is a routine process including data collection, feature engineering, model training, hyper-parameters tuning, model evaluation and model deployment. The process is usually complex, iterated and time-consuming. Commonly, researchers seldom start building the machine model from scratch. They may select some well-known and well-trained models in similar task domains as the reference models. Then they try to tune the hyper-parameters and accelerate the iteration. Thus, some models are often reused and need to be reproduced by using new training dataset. Moreover, understanding the model and the iteration is more necessary. This scenario is very similar to that of software reuse. In this poster, we propose Modelware and argue the need of Modelware to make the machine learning model reusable and reproducible. We define the Modelware which is the reused object and develop a model repository to provide the model lineage management and model visit tool. The big data for building model is managed collaboratively so that the model can be reproduced. The iteration process to obtain the final optimized model is abstracted and implemented using a lightweight workflow. Finally, we take two different classification tasks as the demonstration.

In this paper, we propose a novel weakly supervised semantic segmentation (WSSS) method that uses image tags as supervision to achieve joint pixel-level localization of the key local structure (KLS) and image-level classification of the aurora images captured by the ground-based optical all-sky imager. First, a patch-scale model (PSM) based on the small-scale structure of aurora is designed to identify the type-specific regions for each training image. Second, a region-scale model is trained with the identified type-specific regions to coarsely localize the KLS from multiple sizes of field of view, based on which the aurora image is classified. Finally, given the predicted image type, the PSM further refines the KLS in a pixel level. By localizing KLS from coarse to fine, the proposed method captures both overall shape with a bottom-up processing and local structure details of aurora in a top-down manner. Extensive experiments on the expert labeled data sets have demonstrated the efficacy of the proposed method in benchmarking with the state-of-the-art WSSS methods.

A conventional matrix-based visual secret sharing scheme has the drawbacks of pixel expansion and the requirement of predetermined sophisticated codebooks. A random grid-based visual secret sharing (RGVSS) scheme is an effective approach to solving these two problems. However, up to now, most of the existing publications about the RGVSS scheme deal with the threshold access structures, while there is hardly any appropriate method to construct the RGVSS scheme for general access structures (GASs). In this paper, a novel method to construct a generalized RGVSS (GRGVSS) scheme for GASs is proposed. The construction algorithm consists of two parts. In the first part, a more general (n, n)-GRGVSS scheme is proposed; in addition, the visual quality of the reconstructed secret image for this GRGVSS scheme is formally analyzed. In the second part, we utilize this (n, n)-GRGVSS to construct the GRGVSS scheme for given GASs by treating the procedure as a nonlinear 0-1 programming model. In the experimental phase, by changing the given preconditions, we analyze the security of the proposed scheme and assess the visual quality of the recovered secret images for the proposed scheme under different situations. The simulation results show that the proposed GRGVSS scheme for GASs is feasible and efficient.

Local energy pattern, a statistical histogram-based representation, is proposed for texture classification. First, we use normalized local-oriented energies to generate local feature vectors, which describe the local structures distinctively and are less sensitive to imaging conditions. Then, each local feature vector is quantized by self-adaptive quantization thresholds determined in the learning stage using histogram specification, and the quantized local feature vector is transformed to a number by N-nary coding, which helps to preserve more structure information during vector quantization. Finally, the frequency histogram is used as the representation feature. The performance is benchmarked by material categorization on KTH-TIPS and KTH-TIPS2-a databases. Our method is compared with typical statistical approaches, such as basic image features, local binary pattern (LBP), local ternary pattern, completed LBP, Weber local descriptor, and VZ algorithms (VZ-MR8 and VZ-Joint). The results show that our method is superior to other methods on the KTH-TIPS2-a database, and achieving competitive performance on the KTH-TIPS database. Furthermore, we extend the representation from static image to dynamic texture, and achieve favorable recognition results on the University of California at Los Angeles (UCLA) dynamic texture database.

The naturally occurring aurora phenomenon is a dynamically evolving process. Taking temporal information into consideration, the auroral image sequence analysis is more reasonable and desirable than using static images only. However, the enormous richness of space structures and temporal variations make automatic auroral sequence analysis a particularly challenging task. In this paper, a hidden Markov model (HMM) based representation method including features of spatial texture and dynamic evolution is presented to characterize auroral image sequences captured by all-sky imagers (ASIs). The uniform local binary patterns are employed to describe the 2-D space structures of ASI images. HMM is feasible to characterize the doubly stochastic process involved in the auroral evolution-measurable polar light activities and hidden dynamic plasma processes. We present an affine log-likelihood normalization technique to manage the sequences with different lengths. The proposed method is used in the automatic recognition of four primary categories of ASI auroral observations between the years 2003 and 2009 at the Yellow River Station, Ny-Ålesund, Svalbard. The supervised classification results on manually labeled data in 2003 demonstrate the effectiveness of the proposed technique. Compared with frame-based classification, the higher accuracies and the lower rejection rates show the advantages of the sequence-based method. The occurrence distributions of the four aurora categories were obtained through automatic classification of data gathered from 2004 to 2009. Their agreement with the multiple-wavelength intensity distribution of the dayside aurora and the conclusions made from the frame-based method further illustrate the validity of our method on auroral representation and classification.

In in vivo optical projection tomography (OPT), object motion will significantly reduce the quality and resolution of the reconstructed image. Based on the well-known Helgason-Ludwig consistency condition (HLCC), we propose a novel method for motion correction in OPT under parallel beam illumination. The method estimates object motion from projection data directly and does not require any other additional information, which results in a straightforward implementation. We decompose object movement into translation and rotation, and discuss how to correct for both translation and general motion simultaneously. Since finding the center of rotation accurately is critical in OPT, we also point out that the system's geometrical offset can be considered as object translation and therefore also calibrated through the translation estimation method. In order to verify the algorithm effectiveness, both simulated and in vivo OPT experiments are performed. Our results demonstrate that the proposed approach is capable of decreasing movement artifacts significantly thus providing high quality reconstructed images in the presence of object motion.

Traditional minutiae matching algorithms assume that each minutia has the same discriminability. However, this assumption is challenged by at least two facts. One of them is that fingerprint minutiae tend to form clusters, and minutiae points that are spatially close tend to have similar directions with each other. When two different fingerprints have similar clusters, there may be many well matched minutiae. The other one is that false minutiae may be extracted due to low quality fingerprint images, which result in both high false acceptance rate and high false rejection rate. In this paper, we analyze the minutiae discriminability from the viewpoint of global spatial distribution and local quality. Firstly, we propose an effective approach to detect such cluster minutiae which of low discriminability, and reduce corresponding minutiae similarity. Secondly, we use minutiae and their neighbors to estimate minutia quality and incorporate it into minutiae similarity calculation. Experimental results over FVC2004 and FVC-onGoing demonstrate that the proposed approaches are effective to improve matching performance.

This letter develops a novel method for fingerprint singular point detection based on a new singularity representation of ridge-valley region called orientation entropy. The candidate singular point is obtained by the multiple-scale analysis of orientation entropy and some post processing steps are proposed to filter the spurious core and delta points. An iteration compensation scheme is proposed to search the precise location for core points against the offset further. Performance of the proposed method has been evaluated on the dataset of FVC2002 DB1. Experimental results show that the multiple-scale orientation entropy is correct and effective for singular detection and the location compensation scheme reduces the distance between the detection result and the truth singular point.

In this paper, a logistic dynamic texture model (LDT) is proposed to characterize binary image sequences. Dynamic texture model (DT) is one of the most efficient and successful methods in modeling dynamic sequences. It learns the parameters through a closed-form solution and commonly uses principal component analysis (PCA) to obtain the observation function. PCA assumes a Gaussian distribution over a set of observations. However, the binary image sequences subject to Bernoulli distribution. The LDT introduces logistic PCA to learn the observation function. The proposed model is capable of describing the binary image sequences accurately by processing the pixels of 1 and 0 separately. The model is demonstrated by image reconstructing and activity/gait recognition experiments. Experimental results illustrate the effectiveness of our model.

Information fusion offers a promising solution to the development of a high-performance classification system. In this paper, the problem of multiple gait features fusion is explored with the framework of the factorial hidden Markov model (FHMM). The FHMM has a multiple-layer structure and provides an alternative to combine several gait features without concatenating them into a single augmented feature. Besides, the feature concatenation is used to directly concatenate the features and the parallel HMM (PHMM) is introduced as a decision-level fusion scheme, which employs traditional fusion rules to combine the recognition results at decision level. To evaluate the recognition performances, McNemar's test is employed to compare the FHMM feature-level fusion scheme with the feature concatenation and the PHMM decision-level fusion scheme. Statistical numerical experiments are carried out on the Carnegie Mellon University motion of body and the Institute of Automation of the Chinese Academy of Sciences gait databases. The experimental results demonstrate that the FHMM feature-level fusion scheme and the PHMM decision-level fusion scheme outperform feature concatenation. The FHMM feature-level fusion scheme tends to perform better than the PHMM decision-level fusion scheme when only a few gait cycles are available for recognition.