Effectively selecting an optimal feature subset in high-dimensional data modeling remains a critical challenge. Consequently, feature selection is frequently formulated as a complex multi-objective optimization problem and addressed using multi-objective evolutionary algorithms (MOEAs). However, conventional MOEAs often struggle to capture the complex, underlying interactions among features during the search process. To overcome this limitation, this paper proposes a novel Feature Association Rule Mining-based MOEA (FARM-EA). Leveraging the capacity of association rule mining to reveal strong correlations, co-occurrence patterns, and potential redundancies among features, the proposed algorithm captures the discrepancies in feature selection behaviors between elite and inferior individuals. Subsequently, both synergistic and detrimental feature interactions can be extracted to explicitly guide offspring reproduction. The proposed approach is rigorously evaluated across 18 benchmark classification datasets and compared against five state-of-the-art MOEAs. Experimental results demonstrate that FARM-EA achieves highly competitive performance in terms of both hypervolume (HV) and minimum classification error rate (MCER) metrics.

Object detection in remote sensing images is a critical technology for ensuring urban and environmental safety. However, traditional deep learning methods often struggle with poor detection accuracy, high miss rates, and high deployment costs in complex scenes with multi-scale objects. To address these challenges, we propose EViT-Net, a lightweight network inspired by Vision Transformer (ViT). It significantly improves the efficiency of multi-scale feature processing. First, CA-RepViT is introduced as the backbone network to better extract global features and long-range dependencies. Second, the Generalized Spatial Pyramid Pooling Fast (GSPPF), C3G2, and Lightweight Space-to-Depth (LSPD) components are designed to enhance multi-scale feature fusion while reducing computational costs. Additionally, P2-BiFPN is constructed as the neck network, improving feature representation through comprehensive cross-layer information interaction. Finally, EDetect is developed to optimize the classification and localization capabilities of the detection head, further enhancing detection efficiency. Compared to YOLOv11, EViT-Net reduces the number of parameters and computational load by 56.3% and 12.5%, respectively. On the VisDrone2019 and AI-TOD datasets, EViT-Net improves mean Average Precision (mAP) by 1.8% and 3.9%, recall rate by 2.0% and 5.4%, and Frames Per Second (FPS) by 49.3%. The model's reliability and generalization capability are further validated on the SIMD dataset. Experimental results demonstrate that EViT-Net successfully achieves a balance between detection accuracy and computational efficiency, offering a novel approach for advancing remote sensing image analysis.

Self-supervised learning (SSL) has shown strong potential for learning visual representations from unlabeled data, but transferring these representations effectively to object detection remains challenging. Existing SSL paradigms have complementary strengths and weaknesses: contrastive learning provides strong instance discrimination but lacks spatial sensitivity, while masked image modeling preserves spatial structure but is less effective at distinguishing instances. As a result, current methods often perform well on either classification or detection, but struggle to balance both. To address this problem, we propose Contrastive Masked Histogram-Decoupled Detector (CMHD), a dual-branch SSL framework that combines masked spatial reasoning with gradient-aware contrastive learning. The online branch improves spatial understanding by reconstructing masked regions, while the target branch enhances structural and discriminative representations through gradient-based features. We further introduce a random coordinate attention module to fuse cross-branch features and improve multi-scale representation learning. Extensive experiments on MS COCO, PASCAL VOC, ImageNet-1K, and other benchmarks show that CMHD achieves more balanced performance across classification and detection tasks, outperforming representative pure contrastive learning and masked image modeling baselines. Ablation studies further confirm the effectiveness of each component in the proposed framework.

Objective optimization and constraint satisfaction are two primary and conflicting tasks in solving constrained multi-objective optimization problems (CMOPs). To better trade off them, this paper proposes a two-stage bidirectional coevolutionary algorithm, termed C-TBCEA, for constrained multi-objective optimization. It consists of two stages, with each concentrating on specific targets, i.e., the first stage primarily focuses on objective optimization while the second stage focuses on constraint satisfaction by employing different evolutionary strategies at each stage. Via the synergy of the two stages, a dynamic trade-off between objective optimization and constraint satisfaction can be achieved, thus overcoming the distinctive challenges that may be encountered at different stages of evolution. In addition, to take advantage of both feasible and infeasible solutions, we employ two populations, i.e., the main population that stores the non-dominated feasible solutions and the archive population that maintains the informative infeasible solutions, to prompt the bidirectional coevolution of them. To validate the effectiveness of the proposed C-TBCEA, experiments are carried out on 6 CMOP test suites and 17 real-world CMOPs. The results demonstrate that the proposed algorithm is very competitive with 9 state-of-the-art constrained multi-objective optimization evolutionary algorithms (CMOEAs).

An improved DF-YOLOv8 algorithm was proposed to solve the problem of decreasing target detection accuracy under the influence of occlusion, small target and other factors during target detection on complex road background. FIoU calculated the diagonal distance of the overlap area of the IoU of the real box and the predicted box. Compared with the center point distance of CIoU, the diagonal distance is more suitable for the calculation of the IoU Loss function of the occluded target. The CIoU in the algorithm was replaced by FIoU to improve the regression accuracy of the algorithm for the occluded target. A multi-scale convolution module DCD is proposed for small target problem. The module uses convolution of different scales to extract features of different scales, and fuses channel features of four convolution channels to exchange channel information. To obtain a more comprehensive and rich feature representation. A multi-scale convolutional network is constructed by replacing common convolution in C2f with DCD module to improve the effect of small target detection. The experimental results were compared and verified on the KITTI target detection dataset, and finally the ablation experiment was conducted and compared with the current mainstream detection algorithms, which fully proved the effectiveness of DF-YOLOv8.

This paper aims to propose a trajectory optimization algorithm (TOA) in Mobile Edge Computing (MEC) based on multi-unmanned Aerial Vehicle (UAV) assistance to improve the efficiency of UAV trajectory calculation. This paper analyzes the existing problems in the current MEC and proposes a multi-UAV collaborative computing task processing model. The model optimizes the UAV path planning by designing a TOA and introduces greedy strategies to make the UAV more efficient in completing the task. In the algorithm design process, the intelligent optimization algorithm of the genetic algorithm and particle swarm algorithm is used to optimize the path planning of the UAV and adjust it based on the actual scene data. Meanwhile, the performance and efficiency of the proposed algorithm based on multi-UAV assistance and the traditional random TOA (TRTOA) are compared to evaluate the performance of the proposed TOA. The results show that the proposed TOA based on multi-UAV assistance performs better under different numbers of UAVs. When there are four UAVs, the multi-UAV-assisted TOA proposed here saves an average of 35% of the time and improves computational efficiency by 40% compared with the TRTOA. With six UAVs, the multi-UAV-assisted TOA proposed here saves an average of 45% of the time and improves computational efficiency by 50% compared with the TRTOA. In summary, the TOA based on intelligent optimization and greedy strategy proposed here can effectively enhance the computational efficiency in multi-UAV-assisted MEC, and the research has practical application value.

The intelligent processing of rock data is critical in geological exploitation, where time-sensitive engineering demands pose significant challenges. In the context of big data, handling high-resolution and multipolarized imaging requires efficient and adaptive frameworks. To address the shortcomings of traditional linear attention, we propose IFS-attention, a compact mechanism integrating intuitionistic fuzzy representation and dynamic channel weighting. This approach enhances feature representation by capturing local correlations, improving attention matrix rank, and dynamically prioritizing salient features. Incorporated into transformer architectures, IFS-attention is applied to multipolarized rock thin-section image tasks, such as registration and fusion, achieving state-of-the-art performance while preserving computational efficiency. By addressing the dual challenges of high-resolution imaging and data-intensive geological analysis, this research not only provides a robust solution for multipolarized imaging tasks but also establishes a foundation for integrating lightweight attention within broader computer engineering applications.

Large-scale multi-objective optimization problems (LSMOPs), characterized by hundreds or thousands of decision variables, present significant challenges for existing multi-objective evolutionary algorithms (MOEAs), which often struggle to balance the trade-off among diversity, convergence, and scalability in the vast decision spaces within limited computational resources. To address these challenges, this paper introduces a novel large-scale MOEA that incorporates two key innovations: the adaptive sorting mutation (ASM) and meta-knowledge-assisted sampling (MKAS) strategies. The ASM adaptively perturbs variable values through ascending and descending sorting mutations to enrich the population’s diversity and enhance the exploration of the search space. Meanwhile, the MKAS employs a neural network to learn and predict promising meta-knowledge, thus to dynamically guide the evolutionary efforts towards the Pareto set. Additionally, the competitive swarm optimizer is integrated to further refine the final population, providing a robust baseline for optimization. The performance of the proposed method is evaluated through extensive experiments on 325 LSMOPs with up to 10,000 decision variables and 9 objectives. The results demonstrate the superior convergence, diversity, and scalability of the proposed method compared to existing state-of-the-art algorithms, highlighting its potential as an effective solution for large-scale and complex multi-objective optimization tasks.

Artificial neural networks(ANNs)have made great strides in the field of remote sensing image object detection.However,low detection efficiency and high power consumption have always been significant bottlenecks in remote sensing.Spiking neural networks(SNNs)process information in the form of sparse spikes,creating the advantage of high energy efficiency for computer vision tasks.However,most studies have focused on simple classification tasks,and only a few researchers have applied SNNs to object detection in natural images.In this study,we consider the parsimonious nature of biological brains and propose a fast ANN-to-SNN conversion method for remote sensing image detection.We establish a fast sparse model for pulse sequence perception based on group sparse features and conduct transform-domain sparse resampling of the original images to enable fast perception of image features and encoded pulse sequences.In addition,to meet accuracy requirements in relevant remote sensing scenarios,we theoretically analyze the transformation error and propose channel self-decaying weighted normalization(CSWN)to eliminate neuron overactivation.We propose S3Det,a remote sensing image object detection model.Our experiments,based on a large publicly available remote sensing dataset,show that S3Det achieves an accuracy performance similar to that of the ANN.Meanwhile,our transformed network is only 24.32％as sparse as the benchmark and consumes only 1.46 W,which is 1/122 of the original algorithm's power consumption.

Infrared and visible image fusion strives to create images with enhanced information by seamlessly integrating complementary data from two distinct modalities. However, current progressive image fusion methods encounter a challenge where visible texture details, infrared salient targets, and low-light background information often blend into each other, compromising the clarity of texture and target details. To tackle this challenge, we introduce a progressive two-stage infrared and visible image fusion network focused on enhancing target and texture information, named P3TFusion. In the initial fusion stage, we devise a progressive multi-feature feedback adjustment network (PMFANet) that adaptively integrates dual-branch feature information and performs feedback adjustment to optimize the extraction of texture and target features, ensuring the preservation of more beneficial information. In the subsequent fusion stage, we incorporate a novel embedded texture preference enhancement network (ETPENet), which specializes in capturing fine-grained information of texture preferences by reusing the visible image while effectively suppressing the interference generated by infrared and visible images captured under varying lighting environments. Ultimately, by fine-tuning the loss function, we guarantee that the fused image preserves optimal texture information, color distribution, contrast, and brightness. P3TFusion has been rigorously trained and tested on the LLVIP dataset and further subjected to generalization experiments on the TNO, MSRS, and RoadScene datasets. Compared to state-of-the-art methods, our approach yields superior quantitative results across key evaluation metrics (EN, AG, SF, SD) and stands out in qualitative assessments for its ability to preserve intricate texture and target details.

In the field of Small Object Detection (SOD), accurate classification and localization are crucial for detection performance. However, the inherent imbalance between classification and localization tasks can generate conflicting priorities, leading to suboptimal task coordination for small object detection. This imbalance is mainly caused by the different attention regions and the gradient competition between the two tasks during joint training. In this paper, we propose a Dual-Task Harmonization Framework (DTHF): First, we introduce a Feature Fusion-based Data Augmentation strategy (FF-DA), which amplifies boundary-aware patterns for localization while preserving critical semantic regions for classification, thereby aligning their region-of-interest priorities. Second, we design a Gradient Equilibrium Module (GEM) that dynamically balances tasks by altering the gradients, preventing one task from overwhelming the other during optimization. Experiments on the MS COCO and VisDrone datasets demonstrate that our method, compared to the baseline model, the experimental metrics of our method mAP in the VisDrone data set are improved by 2.0+

In the field of Small Object Detection (SOD), accurate classification and localization are crucial for detection performance. However, previous works have largely focused on region accuracy, with less attention given to spatial information. It makes the performance of the detector in small objects constrained. Therefore, we propose a novel object detection algorithm named space-aware small object detector (SASOD). Specifically, to enrich spatial features during the feature extraction process, we designed a Spatial-aware Convolution module (SAConv), which effectively captures and reconstructs the spatial structural information of images. Concurrently, we enhanced the model’s boundary perception capability by introducing a structural similarity penalty term in the loss function. In the experimental section, we employed the widely-used VisDrone2019 dataset to evaluate the performance of the SASOD. YOLOv10 is selected as the benchmark experiment, and the designed algorithm is fused to YOLOv10, compared with the baseline model, the fused model’s experimental metrics mAP on the VisDrone2019 dataset is improved by 2.46

With the robust development of artificial intelligence (AI), especially image processing has made information technology more efficient and effective in the sense to evaluate facial features, even though there has been masked on the face. However, the accuracy of acne detection and related severity analysis is becoming a significant prospect for the precise treatment of patients. Due to this, close severity is one of the features that need to be added first, while it is considered a highly challenging aspect for dermatologists because the similar appearance of acne in the face reduces the rate of accuracy when examining. It poses a serious problem in the domain of biomedical processing and controls. In this paper, we contribute to four different folds. Initially, this paper presents a novel framework that provides a platform in order to measure localization and segmentation. In this process, consultants receive better accuracy and efficiency during the process of acne detection and severity analysis. Second, this paper utilizes deep neural networks (DNNs) as a backend process to lightning the extraction of multi-scale features through a multi-hierarchy neural net for regionalized facial features to investigate distinction and localization. Third, a class-based segmentation approach customizes and integrates with the proposed framework to examine the background and facial skin separation to distinguish different classes to obtain severity marking. Fourth, the facial skin segmentation classes are built as a cluster segment using generative adversarial network (GAN). With a performance rate of 3.112

Traffic accidents remain a critical issue that significantly impacts public safety and poses major challenges to intelligent transportation systems. The integration of Unmanned Aerial Vehicles (UAVs) with object detection technology offers a promising solution to this problem. However, existing detection networks often exhibit limitations such as missed detections and inadequate real-time performance, particularly for small vehicle targets in remote sensing images, and fail to meet the efficiency requirements of edge computing devices. To address these challenges, this study proposes RL-Net, a rapid and lightweight network model based on enhancements to the YOLOv9s architecture. First, MobileNetV4 is introduced to optimize initial feature extraction, significantly improving the network’s efficiency. Second, the Lightweight Spatial Pyramid Pooling Fast (LSPPF) structure is designed to enhance multiscale feature extraction while accelerating computational speed. Additionally, the Lightweight Representation Cross Stage Partial with ELAN (LRepCSPELAN) module is proposed to further reduce the model’s memory and computational resource demands. Finally, an enhanced feature fusion network is designed to improve detection performance for tiny vehicle targets. Comprehensive evaluations on the VisDrone2019 and UA-DETRAC datasets demonstrate that, compared to YOLOv9s model, RL-Net achieves a 34.5

One of the emerging paradigms in the diagnosis and severity assessment of skin disorders, particularly acne, on the face is the use of advanced digital technology (ADT) in skin disease monitoring. It is said to be a very prevalent issue that has to be looked at by experts in this day and age. Nonetheless, the traditional approach to acne diagnosis still relies on the opinions and expertise of medical professionals. There could be fatal outcomes from both delayed and inaccurate diagnoses. Since acne is a condition that directly affects the healthcare system, this study focusses on accelerating the diagnostic process and closing the gap between diagnosis and treatment. In this work, we introduce a smart face acne disease level monitoring device that allows acne sufferers in different geographical locations to track the severity and specifics of their acne and to communicate precautions. Convolutional neural networks, or CNNs, play a major role in this suggested architecture's AI-enabled cloud-based IoT device interconnectivity. Based on a set of photos, CNNs predict the degree of face acne, which could have implications for further study. This proposed study also addresses the influence of age. Geographically speaking, the architecture provides all the domains of skin diagnostic and preventive scheme, notably for acne diagnosis, addressing the present issue faced by patients with limited or no access to e-healthcare services.

Fast sorting of large-scale data is an essential task for data centers. In previous works, the existing computational model of sorting kernel still results in lower bandwidth utilization on the external memory bus. And the execution of merge operations in merge sort circuit on FPGAs depends on control commands from the host CPU. In this case, the merge sort circuit is not fully offloaded to hardware layer for acceleration, resulting in a performance loss. We design an on-chip merge sort controller to efficiently command the merge sort process. The proposed controller has the ability to schedule multiple on-chip computing kernels simultaneously in a more efficient mode, thus ensuring that the circuit has a better bandwidth utilization. Meanwhile, fundamental factors affecting the performance of merge sort are studied and analyzed, and we propose a high-performance merge sort architecture. Results show that using the proposed controller-centered architecture, an overall improvement of 20-30% in sorting throughput can be achieved. Compared with the state-of-the-art previous merge sorting implementation on FPGA, our circuit can achieve 1.22/1.46\(\times\) speedup.

The fusion of multi-polarized petrographic images of rock thin sections involves the fusion of feature information from microscopic images of rock thin sections illuminated under both plane-polarized and orthogonal-polarized light. During the fusion process of rock thin section images, the inherent high resolution and abundant feature information of the images pose substantial challenges in terms of computational complexity when dealing with massive datasets. In engineering applications, to ensure the quality of image fusion while meeting the practical requirements for high-speed processing, this paper proposes a novel fast fusion Transformer. The model leverages a soft matching algorithm based on intuitionistic fuzzy sets to merge redundant tokens, effectively mitigating the negative effects of asymmetric dependencies between tokens. The newly generated artificial tokens serve as brokers for the Query (Q), forming a novel lightweight fusion strategy. Both subjective visual observations and quantitative analyses demonstrate that the Transformer proposed in this paper is comparable to existing fusion methods in terms of performance while achieving a notable enhancement in its inference efficiency. This is made possible by the attention paradigm, which is equivalent to a generalized form of linear attention, and the newly designed loss function. The model has been experimented on with multiple datasets of different rock types and has exhibited robust generalization capabilities. It provides potential for future research in diverse geological conditions and broader application scenarios.

Fast regular expression matching is an essential task for deep packet inspection. In previous works, the regular expression matching engine on FPGA struggled to achieve an ideal balance between resource consumption and throughput. Speculation and enumerative computation exploits the statistical properties of deterministic finite automata, allowing for more efficient pattern matching. Existing related designs mostly revolve around vector instructions and multiple processors/cores or SIMD instruction sets, with a lack of implementation on FPGA platforms. We design a parallelized two-character matching engine on FPGA for efficiently fast filtering off fields with no pattern features. We transform the state transitions with sequential dependencies to the existing problem of elements in one set, enabling the proposed design to achieve high throughput with low resource consumption and support dynamic updates. Results show that compared with the traditional DFA matching, with a maximum resource consumption of 25% for on-chip FFs (74323/1045440) and LUTs (123902/522720), there is an improvement in throughput of 8.08-229.96 × speedup and 87.61-99.56% speed-up(percentage improvement) for normal traffic, and 11.73-39.59 × speedup and 91.47-97.47% speed-up(percentage improvement) for traffic with high-frequency match hits. Compared with the state-of-the-art similar implementation, our circuit on a single FPGA chip is superior to existing multi-core designs.

Constrained multi-objective problems face the chal-lenge of simultaneously optimizing the objective functions and constraint satisfaction. The difficulty in addressing this challenge lies in considering convergence, feasibility, and diversity simulta-neously. To better solve CMOPs, this paper proposes a multi- population constrained multi-objective evolutionary algorithm based on knowledge transfer (C-MTEA). It consists of three different populations, i.e., the main population, the archive population, and the auxiliary population, that can cooperate with each other to evolve collectively. Specifically, the main population and the archive population cooperate by utilizing different search strategies to generate complementary offspring, while the auxiliary population, which does not consider con-straints, can assist the main population in convergence. In the environmental selection stage, useful information is transferred across populations by sharing offspring generated by various strategies, thus facilitating the evolution of populations. To validate the effectiveness of the proposed C- MTEA, experiments are carried out on 5 popular benchmark suites containing up to 63 instances. The results demonstrate that the proposed algorithm is competitive with state-of-the-art constrained multi-objective optimization evolutionary algorithms (CMOEAs).