Recent advances in medical image segmentation have significantly improved segmentation accuracy. Nevertheless, the clinical deployment of large-scale segmentation networks remains constrained by challenges such as excessive parameter counts, complex architectures, and limited adaptability to diverse deployment environments. The absence of lightweight design further restricts their integration into resource-limited edge devices. To address these barriers, lightweight strategies have emerged as an effective solution. Structural optimization simplifies network architectures to reduce computational costs, while model compression techniques shrink model size without sacrificing performance. At the same time, hardware-level acceleration provides additional support for efficient inference in real-world scenarios. This review systematically summarizes recent lightweight methods for medical image segmentation from both software and hardware perspectives. Representative algorithmic approaches are highlighted, including pruning, quantization, knowledge distillation, and efficient network architectures, along with hardware-aware optimization strategies tailored for edge deployment. Moreover, we explored the mainstream approach of integrating large-scale models with lightweight technologies to achieve the optimal balance between segmentation accuracy and computational efficiency. Finally, current limitations and potential research directions are outlined to promote the translation of lightweight segmentation models into routine clinical workflows. By providing a structured reference, this review aims to support researchers and practitioners in advancing the efficient and practical application of medical image segmentation in clinical environments.

Visible light and thermal infrared tracking combines the characteristics of visible light and thermal infrared modalities to achieve robust target tracking in all-weather and all-day scenarios. However, most existing visible light and thermal infrared tracking methods rely on either full fine-tuning or attention mechanisms, which introduce a large number of parameters and are predominantly influenced by the visible modality. This results in challenges such as high computational complexity, slower processing speeds, and limited exploitation of multimodal information. To address these issues, this paper proposes a lightweight multimodal tracking model based on feature fusion and enhancement. The model consists of a feature fusion adapter and a joint enhancement adapter, designed to integrate and refine information across modalities. It employs a dual-stream transformer encoder with shared parameters across modality branches, utilizing a frozen pre-trained foundation model to independently extract features from visible light and thermal infrared inputs. The lightweight fusion adapter combines modality-specific information, while the joint enhancement adapter refines unimodal features, introducing only 0.23M trainable parameters. Experimental results on the LasHeR benchmark demonstrate that the proposed method outperforms prompt learning and other adapter-based methods, achieving a 4.4% improvement in PR and a 3.3% increase in SR while maintaining computational efficiency. With a real-time inference speed of 28.60 FPS, the proposed method balances accuracy and efficiency effectively. The source code will be available at https://github.com/huxue/MFJA.

In recent years, federated learning (FL) has received widespread attention for its ability to enable collaborative training across multiple clients while protecting user privacy, especially demonstrating significant value in scenarios such as medical data analysis, where strict privacy protection is required. However, most existing FL frameworks mainly focus on data heterogeneity without fully addressing the challenge of heterogeneous model aggregation among clients. To address this problem, this article proposes a novel FL framework called FedMKD. This framework introduces proxy models as a medium for knowledge sharing between clients, ensuring efficient and secure interactions while effectively utilizing the knowledge in each client's data. In order to improve the efficiency of asymmetric knowledge transfer between proxy models and private models, a hybrid feature-guided multilayer fusion knowledge distillation (MKD) learning method is proposed, which eliminates the dependence on public data. Extensive experiments were conducted using a combination of multiple heterogeneous models under diverse data distributions. The results demonstrate that FedMKD efficiently aggregates model knowledge.

Small-object detection remains challenging due to limited pixel information, blurred boundaries, and weak semantic cues. Although recent advances in multiscale fusion and attention mechanisms have led to improved performance, existing methods still struggle to preserve high-frequency structural details and achieve precise localization-particularly in dense, cluttered, or low-resolution scenarios. These limitations are primarily caused by the loss of fine-grained features during downsampling and the absence of region-aware focus mechanisms. Inspired by the human visual strategy of "see clearly and detect precisely," we propose PEFC-Net, a novel framework that enhances both perceptual clarity and localization accuracy for small-object detection. To mitigate structural degradation, we introduce the hybrid structural perception (HSP) module, which jointly encodes spatial gradients and localized frequency components through wavelet-based decomposition and edge-aware refinement. To further improve region-level focus, we design the axis-aligned focus calibration (AAFC) module, which captures long-range directional context via axis-sensitive pooling and adaptively refines attention with shape-aware calibration. Extensive experiments on four challenging benchmarks-VisDrone-2019, TT100K, NWPU VHR-10, and DIOR-demonstrate that PEFC-Net consistently outperforms state-of-the-art methods, delivering robust performance under occlusion, dense distribution, and scale variation.

Heterogeneous federated learning enables collaborative training across clients under dual heterogeneity of models and data, posing challenges for effective knowledge transfer. Federated mutual learning employs proxy models to bridge cross-model knowledge exchange; however, existing methods remain limited to direct alignment between the outputs of private and proxy models, ignoring the deep discrepancies in representation and decision spaces between them. Such cognitive biases cause knowledge to be transferred only at shallow levels and trigger performance bottlenecks. To address this, this paper proposes FedKWAZ to identify and exploit Knowledge Weak-Aware Zones (KWAZ)—spatial zones of deep knowledge misalignment between private and proxy models, further refined into Semantic Weak-Aware Zones and Decision Weak-Aware Zones, which characterize cognitive misalignments in representation and decision spaces as focal targets for enhanced bidirectional distillation. FedKWAZ designs a Hierarchical Adaptive Patch Mixing (HAPM) mechanism to generate multiple mixed samples and employs a Knowledge Discrepancy Perceptron (KDP) to select the samples exhibiting the largest representation and decision discrepancies, thereby mining critical KWAZ. These modules are integrated into a two-stage mutual learning framework, achieving global class-level representation-decision consistency alignment and local KWAZ-guided refinement, structurally bridging cognitive biases across heterogeneous mutual learning models. Experimental results on multiple datasets and model configurations demonstrate the superior performance of FedKWAZ.

Driver distraction detection is critical to reducing road traffic accidents and increasing the efficiency of advanced driver assistance systems. Real-time lightweight models are especially important for in-vehicle devices with limited computing resources. However, most existing methods focus on designing lighter network architectures and ignore the performance loss when detecting tiny targets. In order to realize the collaborative optimization of tiny target detection accuracy and network lightweight, a driver distraction detection method ATD 2 Net based on adaptive tiny target detection and lightweight networks is proposed. This method aims to reduce model complexity while fully capturing target features for accurate detection. ATD 2 Net consists of three core modules, Channel Reconstruction Perception Module (CRPM), Dynamic Spatial Self-locking Module (DSSM) and Structural Feedback Optimization Module (SFOM). CRPM reconfigures channels and reconstructs them into batch dimensions, uses parallel strategies to perceive interactive features between channels, and significantly enhances feature extraction capabilities. DSSM adopts dynamic locking and adaptive spatial selection mechanisms to capture multi-scale features while injecting adaptive spatial information. It effectively aggregates instance features and reduces the interference of conflicting information and background information, thereby improving the detection ability of tiny targets. SFOM uses dependency trees to model inter-layer relationships and integrate coupling parameters into groupings. It uses a sparse strategy to remove unimportant parameters, achieving lightweight modeling while balancing accuracy and speed. Experimental results show that ATD 2 Net is superior to the latest methods in driver distraction detection, showing excellent performance and good application prospects.

Object detection in remote sensing images is crucial in numerous fields; however, it becomes highly challenging under adverse weather circumstances. Given that previous remote sensing image object detection methods were designed based on normal weather conditions and ideal datasets, they are not beneficial for detection under real-world haze conditions and with class-imbalanced data. In this work, an adaptive dehazing centroid contrastive network (DHC-Net) is proposed to address the aforementioned issues. This network consists of an adaptive dehazing module and a centroid-guided contrastive learning approach. The adaptive dehazing module learns the image content to generate adaptive dehazing parameters, thus alleviating the influence of haze on the quality of remote sensing images. The centroid-guided contrastive learning approach is particularly designed to address the issue of imbalanced datasets. Integrating centroid vectors with actual samples in each training batch guarantees that each class is sampled at least once, effectively preventing the undersampling of minority classes. Moreover, dynamic weighted sampling based on prediction confidence guides the model to give priority to smaller classes, remarkably improving its ability to handle imbalanced data. Extensive experiments on the DOTA-v2.0, DOTA-v2.0Haze, RTTS, and HazeNet datasets demonstrate that DHC-Net is outstanding in handling haze conditions in remote sensing data, substantially enhancing target detection accuracy, even in the presence of imbalanced object classes. The source code will be available athttps://github.com/Linghuaqian1/DHC_Net

In this article, we propose a feature distillation-based uniformity few-shot domain adaptation (FUFD), for cross-domain fault diagnosis with sample shortage. To address the the few-shot problem, a uniformity prototypical contrastive network is designed to improve the data sensitivity of the model. Compared to the vanilla prototypical network, the learned prototypes contain more information about fault classes by encoding semantic structure information into the feature space while dynamically estimating the distribution concentration around each class prototype. Uniformity and correlation principles are introduced to alleviate prototype collapse: the uniformity principle ensures balanced prototype distribution, while the correlation principle enhances the diversity and distinctiveness of prototypical features. In addition, a cross-domain feature distillation-based domain adaptation module is designed to address significant domain shift. This module softens the class-specific information to capture more domain-consistent information and avoid overfitting to source working condition. Finally, experiments and ablation studies on cross-domain bearing fault diagnosis tasks with limited samples validate the effectiveness of FUFD and its individual modules in enhancing few-shot cross-domain fault diagnosis performance.

A certainty and transferability guided few-shot domain adaptation network is proposed to address few-shot open-set cross-domain fault diagnosis in this article. The proposed method is composed of a feature extractor, a certainty-guided prototypical contrastive module and a transferability weighting domain adaptation module. The certainty-guided prototypical contrastive module based on samples informative importance is designed to enhance the data sensitivity with limited samples while achieving well class separation for open-set scenarios. The module infers informative importance of samples to guide method learn more effective representations. Meanwhile, correlation and uniformity principles are incorporated to alleviate prototype collapse. The transferability weighting domain adaptation module is designed to address great domain gaps and negative transfer caused by asymmetrical label spaces. The module quantifies sample transferability and down-weights the irrelevant samples based on their transferability scores. Experimental results on few-shot open-set cross-domain bearing fault diagnosis tasks demonstrated the superior and effectiveness of the proposed method.

Accurate 3D medical image segmentation is crucial for diagnosis and treatment. Diffusion models demonstrate promising performance in medical image segmentation tasks due to the progressive nature of the generation process and the explicit modeling of data distributions. However, the weak guidance of conditional information and insufficient feature extraction in diffusion models lead to the loss of fine-grained features and structural consistency in the segmentation results, thereby affecting the accuracy of medical image segmentation. To address this challenge, we propose a Mamba-Enhanced Diffusion Model for 3D Medical Image Segmentation. We extract multilevel semantic features from the original images using an encoder and tightly integrate them with the denoising process of the diffusion model through a Semantic Hierarchical Embedding (SHE) mechanism, to capture the intricate relationship between the noisy label and image data. Meanwhile, we design a Global-Slice Perception Mamba (GSPM) layer, which integrates multi-dimensional perception mechanisms to endow the model with comprehensive spatial reasoning and feature extraction capabilities. Experimental results show that our proposed MambaDiff achieves more competitive performance compared to prior arts with substantially fewer parameters on four public medical image segmentation datasets including BraTS 2021, BraTS 2024, LiTS and MSD Hippocampus. The source code of our method is available at https://github.com/yuliu316316/MambaDiff.

3D medical images are volumetric data that provide spatial continuity and multi-dimensional information. These features provide rich anatomical context. However, their anisotropy may result in reduced image detail along certain directions. This can cause blurring or distortion between slices. In addition, global or local intensity inhomogeneities are often observed. This may be due to limitations of the imaging equipment, inappropriate scanning parameters, or variations in the patient's anatomy. This inhomogeneity may blur lesion boundaries and may also mask true features, causing the model to focus on irrelevant regions. Therefore, a probability map-guided network for 3D volumetric medical image segmentation (3D-PMGNet) is proposed. The probability maps generated from the intermediate features are used as supervisory signals to guide the segmentation process. A new probability map reconstruction method is designed, combining dynamic thresholding with local adaptive smoothing. This enhances the reliability of high-response regions while suppressing low-response noise. A learnable channel-wise temperature coefficient is introduced to adjust the probability distribution to make it closer to the true distribution; in addition, a feature fusion method based on dynamic prompt encoding is developed. The response strength of the main feature maps is dynamically adjusted, and this adjustment is achieved through the spatial position encoding derived from the probability maps. The proposed method has been evaluated on four datasets. Experimental results show that the proposed method outperforms state-of-the-art 3D medical image segmentation methods. The source codes have been publicly released at https://github.com/ZHANGZIMENG01/3D-PMGNet.

In the field of object detection, detecting small objects is an important and challenging task. However, most existing methods tend to focus on designing complex network structures, lack attention to global representation, and ignore redundant noise and dense distribution of small objects in complex networks. To address the above problems, this paper proposes a small object detection method based on global multi-level perception and dynamic region aggregation. The method achieves accurate detection by dynamically aggregating effective features within a region while fully perceiving the features. This method mainly consists of two modules: global multi-level perception module and dynamic region aggregation module. In the global multi-level perception module, self-attention is used to perceive the global region, and its linear transformation is mapped through a convolutional network to increase the local details of global perception, thereby obtaining more refined global information. The dynamic region aggregation module, devised with a sparse strategy in mind, selectively interacts with relevant features. This design allows aggregation of key features of individual instances, effectively mitigating noise interference. Consequently, this approach addresses the challenges associated with densely distributed targets and enhances the model’s ability to discriminate on a fine-grained level. This proposed method was evaluated on two popular datasets. Experimental results show that this method outperforms state-of-the-art methods in small object detection tasks, demonstrating good performance and potential applications.

Unsupervised domain adaptation is widely used for fault diagnosis under variable working conditions. However, loss oscillation and slow convergence, which are caused by the dynamically varying alignment of targets during domain adaptation, are ignored. Therefore, a Gaussian mixture variational based transformer domain adaptation (GMVTDA) fault diagnosis method is proposed. A feature extractor based on transformer layers is designed to capture long-term dependency information and local features. Subsequently, a domain alignment term is proposed to project the features learned from both working conditions into the common assistance distribution and make them follow the same distribution after the alignment process. Additionally, considering that fault diagnosis is a multiclassification process, a Gaussian mixture is utilized to build the common assistance distribution. Ultimately, the proposed GMVTDA is applied to bearing fault diagnosis under variable working conditions, and the experimental results prove its effectiveness.

Federated learning (FL), as a machine learning method for multi-client model aggregation, faces performance problems caused by data heterogeneity. At the same time, global model aggregation will also cause customers to forget their own personalized knowledge, resulting in poor local training results. To this end, we propose a double distillation personalized federated learning framework (pFedTD) that combines local self-knowledge distillation and global non-ground truth class knowledge distillation. This method effectively balances personalization and global performance by extracting the personalized and global historical knowledge of each client and using a parameter adaptation method to weigh the intensity of self-distillation and global distillation. pFedTD can also perform stably on large-scale heterogeneous data and can alleviate local and global forgetting problems. Our experiments demonstrate that our method outperforms other baselines even on non-IID data.

Due to the continuous improvement of traffic analysis technology, traditional covert channels have become insecure and vulnerable to human sabotage. Blockchain technology has the characteristics of immutability and anonymity, making covert communication more unmonitored and robust. However, it also brings about low communication efficiency. In this article, we adopt the idea of transaction rounds and propose for the first time the construction of HMAC values order (HVO) scheme. Furthermore, we further propose a HMAC values and transaction matrices (HV-TM) scheme to improve communication efficiency. This article is the first to use the gas field to embed data to improve the embedding rate. Use the random numbers generated by the Mersenne Twister algorithm to disrupt the order of addresses to improve the concealment of reused addresses. Experiments have shown that the two schemes have higher communication efficiency and better embedding rate than existing schemes.

Machine learning related research in transient control has drawn considerable attention with the rapid increase in data measurement from power grids. Two key components, the control algorithm and system structure, work together to determine the control performance. The design of control laws, the selection of phase measurement units, the allocation of power resources, and the scheduling of communication topology in limited cyber-physical resources need to be considered. Many existing scheduling or planning schemes specialized for control structure are designed based on various linearized analytical models or the optimization of steady states. However, the transient dynamics of power grids are nonlinear and parts of these dynamics are usually unknown. Linearized analytical models cannot represent the transient dynamics of power grids with large disturbances. This article proposes a sparse neural network based reinforcement learning scheme to optimize the control system structure for the transient stability enhancement of power grids with energy storage systems. One adjustable group sparse weight matrix is introduced to formulate both control structure and actor–critic networks. This strategy enables the proposed scheme to simultaneously schedule the control system structure and design the control laws by online learning without solving any combinational optimization problems or requiring any linearized analytical models. The sufficient conditions of learning stability, control stability, and group sparsity are thoroughly studied by mathematical analysis. The proposed scheme is simulated on an IEEE 118-bus test system for verification. The simulation results confirm the feasibility, advantages, and adaptability of the proposed method.

Abstract Object detection is one of the main tasks in computer vision and has made great progress in recent years. However, the performance of target detectors is significantly dropped by the differences between existing datasets and application scenarios, leading to the so‐called domain shift problem. To address such an issue, a novel co‐teaching based pseudo label refinery framework for cross‐domain object detection is developed, which cooperates with two models to select data from target domain for each other. This strategy can effectively purify the predicted pseudo labels and resist noisy labels. Specifically, the framework consists of two encoders (i.e. structure encoder and global encoder), two classifiers and one discriminator, in which structure encoder is used to extract structural features that are not disturbed by colour, and the global encoder is used to extract the complete discriminant features. The two encoders are each followed by a classifier. In training, the structure and global encoder with labelled source samples are first trained, so that it has the initial recognition ability. Then the samples assigned are used with pseudo labels by the classifier following the structure encoder to fine‐tune the global encoder which pre‐trained on the labelled source domain and obtain the refined labels for the target data. With the refined labels, the structure encoder is further optimised on the target domain. During this process, the proposal is to cross use the two classifiers to promote the mutual transfer of complementary capabilities of the two encoders. Moreover, a novel residual channel attention block (RCA) embedded with salient features is designed to pay more attention to the target regions. Extensive experiments demonstrate that the developed framework can generate clean labels for unlabelled target data and boost the performance of cross domain object detection. The code is available at http://www.msp‐lab.cn:1436/msp/cbplr‐master .

In cloud computing, data is duplicated to prevent data loss. One way to achieve data consistency in such a distributed computing systems is to use a blockchain. Based on practical Byzantine fault tolerance (PBFT), a specific type of blockchain, this paper proposes a synchronous Byzantine fault tolerance (SBFT) algorithm that not only maintains data consistency, but also has much higher efficiency than other general blockchain algorithms. We provide experimental results that demonstrate the algorithm's data consistency, efficiency, and reliability.

A new algebraic system, test algebra (TA), is proposed for identifying faults in combinatorial testing for software-as-a-service (SaaS) applications. In the context of cloud computing, SaaS is a new software delivery model, in which mission-critical applications are composed, deployed, and executed on cloud platforms. Testing SaaS applications is challenging because new applications need to be tested once they are composed, and prior to their deployment. A composition of components providing services yields a configuration providing an SaaS application. While individual components in the configuration may have been thoroughly tested, faults still arise due to interactions among the components composed, making the configuration faulty. When there are k components, combinatorial testing algorithms can be used to identify faulty interactions with t or fewer components, for some threshold 2 ≤ t ≤ k on the size of interactions considered. In general, these methods do not identify specific faults, but rather indicate the presence or absence of some faults. To identify specific faults, an adaptive testing regime repeatedly constructs and tests configurations in order to determine, for each interaction of interest, whether it is faulty or not. In order to perform such testing in a loosely coupled distributed environment such as the cloud, it is imperative that testing results can be combined from many different servers. The TA defines rules to permit results to be combined, and to identify the faulty interactions. Using the TA, configurations can be tested concurrently on different servers and in any order. The TA always keeps the high reduction rate of potential faulty configurations in fault location analysis.