CoroYOLO: a novel colorectal cancer detection method based on the Mamba framework DOI Creative Commons
Wenfei Chen,

Fengrui Hou,

Yue Shen

et al.

Frontiers in Physics, Journal Year: 2025, Volume and Issue: 13

Published: May 22, 2025

Colorectal cancer (CRC) is one of the most common malignant tumors worldwide, and early detection crucial for improving cure rates. In recent years, object methods based on convolutional neural networks (CNNs) transformers have made significant progress in medical image analysis. However, CNNs limitations capturing global contextual information, while can handle long-range dependencies, their high computational complexity limits efficiency practical applications. To address these issues, this paper proposes a novel model—CoroYOLO. CoroYOLO builds upon YOLOv10 architecture by incorporating concept State Space Model (SSM) introduces TSMamblock module, which dynamically models input data, reduces redundant computations, improves both accuracy. Additionally, integrates Efficient Multi-Scale Attention (EMA) mechanism, adaptively strengthens focus critical regions, enhancing model’s robustness complex images. Experimental results show that after training SUN Polyp PICCOLO datasets, outperforms existing mainstream Etis-Larib dataset, achieving state-of-the-art performance demonstrating effectiveness colorectal detection.

Language: Английский

CPLOYO: A pulmonary nodule detection model with multi-scale feature fusion and nonlinear feature learning DOI
Meng Wang, Zi Jian Yang,

Ruifeng Zhao

et al.

Alexandria Engineering Journal, Journal Year: 2025, Volume and Issue: 122, P. 578 - 587

Published: March 20, 2025

Language: Английский

Citations

0
EduVQA: A multimodal Visual Question Answering framework for smart education DOI
J. Xiao, Zipeng Zhang

Alexandria Engineering Journal, Journal Year: 2025, Volume and Issue: 122, P. 615 - 624

Published: March 22, 2025

Language: Английский

Citations

0
MAF-Net: A multimodal data fusion approach for human action recognition DOI Creative Commons
Dongwei Xie, Xiaodan Zhang,

Xiang Gao

et al.

PLoS ONE, Journal Year: 2025, Volume and Issue: 20(4), P. e0319656 - e0319656

Published: April 9, 2025

3D skeleton-based human activity recognition has gained significant attention due to its robustness against variations in background, lighting, and viewpoints. However, challenges remain effectively capturing spatiotemporal dynamics integrating complementary information from multiple data modalities, such as RGB video skeletal data. To address these challenges, we propose a multimodal fusion framework that leverages optical flow-based key frame extraction, augmentation techniques, an innovative of streams using self-attention modules. The model employs late strategy combine features, allowing for more effective capture spatial temporal dependencies. Extensive experiments on benchmark datasets, including NTU RGB+D, SYSU, UTD-MHAD, demonstrate our method outperforms existing models. This work not only enhances action accuracy but also provides robust foundation future integration real-time applications diverse fields surveillance healthcare.

Language: Английский

Citations

0
Deep neural network-based music user preference modeling, accurate recommendation, and IoT-enabled personalization DOI
Jing Lin,

Siyang Huang,

Yujun Zhang

et al.

Alexandria Engineering Journal, Journal Year: 2025, Volume and Issue: 125, P. 232 - 244

Published: April 16, 2025

Language: Английский

Citations

0
Network traffic prediction based on transformer and temporal convolutional network DOI Creative Commons

Yi Wang,

P. Chen

PLoS ONE, Journal Year: 2025, Volume and Issue: 20(4), P. e0320368 - e0320368

Published: April 23, 2025

This paper proposes a hybrid model combining Transformer and Temporal Convolutional Network (TCN). addresses the shortcomings of current approaches in capturing long-term short-term dependencies network traffic prediction tasks. The module effectively captures global temporal relationships through multi-head self-attention mechanism. Meanwhile, TCN models local using dilated convolution technology. Experimental results on PeMSD4 PeMSD8 datasets demonstrate that our method considerably surpasses mainstream methods at all time steps, particularly step prediction. Through ablation experiments, we verified contribution each to performance, further proving key role modules improving performance.

Language: Английский

Citations

0
The Financial Institution Text Data Mining and Value Analysis Model Based on Big Data and Natural Language Processing DOI Open Access

Juan Yang,

Qiang Cai, Jie Gong

et al.

Journal of Organizational and End User Computing, Journal Year: 2025, Volume and Issue: 37(1), P. 1 - 40

Published: April 26, 2025

Financial markets are inherently complex and influenced by a variety of factors, making it challenging to predict trends detect key events. Traditional models often struggle integrate both structured, or numerical, unstructured, textual, data; additionally, they fail capture temporal dependencies the dynamic relationships between financial entities. To address this, multidimensional integrated model for text mining value analysis (MI-FinText), was proposed. MI-FinText multi-task learning, graph convolutional networks knowledge construction. simultaneously performed sentiment analysis, event detection, prediction learning shared representations across tasks modeling time-dependent continuously updated reflect evolving landscape, enabling real-time insights.

Language: Английский

Citations

0
NAH-GNN: A graph-based framework for multi-behavior and high-hop interaction recommendation DOI Creative Commons

Guangzhu Tan

PLoS ONE, Journal Year: 2025, Volume and Issue: 20(4), P. e0321419 - e0321419

Published: April 29, 2025

With the growing demand for personalized marketing, recommender systems have become essential tools to help users quickly discover products or content that match their interests. However, traditional recommendation methods face significant limitations in handling complex user behaviors and sparse data, particularly accurately capturing relationships among diverse interaction types higher-order dependencies. To address these challenges, this paper proposes a novel model based on graph neural networks (MBH-GNN) optimize marketing strategies. MBH-GNN constructs multi-behavior employs neighborhood-aware modeling effectively integrate user-item (e.g., browsing, favoriting, purchasing), dynamically assigning weights generate semantically rich embeddings. Furthermore, incorporates high-hop relational learning mechanism capture long-range dependencies, enhancing its ability contextual information. These features enable achieve higher accuracy diversity scenarios. Experimental results demonstrate significantly outperforms existing baseline methods, achieving HR@10 of 0.789 NDCG@10 0.330 BeiBei dataset, 0.773 0.319 Tmall dataset. The exhibits exceptional robustness adaptability, addressing data sparsity cold-start This study offers an efficient scalable solution providing critical theoretical support practical value improving system performance behavior challenges.

Language: Английский

Citations

0
Integration of smart sensors and phytoremediation for real-time pollution monitoring and ecological restoration in agricultural waste management DOI Creative Commons

Jinsong Guo,

Xiaoxin Lin,

Yi Xiao

et al.

Frontiers in Plant Science, Journal Year: 2025, Volume and Issue: 16

Published: May 13, 2025

Global climate change and ecological degradation highlight the urgency of dealing with agricultural waste restoration. Traditional pollutant monitoring restoration methods face challenges in accuracy adaptability, especially when complex environmental data. This paper proposes Bio-DANN model, which combines biogeochemical models deep learning techniques to improve prediction. The model uses neural networks (DNNs) attention mechanisms process multidimensional data various scenarios real time. Experimental results based on Open Soil Data NEON datasets show that performs well prediction, mean square errors (MSE) 0.012 0.018, root (RMSE) 0.109 0.134, 0.92 0.90, respectively. In terms assessment, achieved Δ F PIPGR 0.15 18%, 0.20 22%, respectively, H’ values 1.5 1.7, are better than other models. provides a promising technical solution for protection, resource recovery sustainable agriculture, showing significant potential monitoring, soil health assessment evaluation.

Language: Английский

Citations

0
Path planning algorithm for logistics autonomous vehicles at Cainiao stations based on multi-sensor data fusion DOI Creative Commons

Yan Chen

PLoS ONE, Journal Year: 2025, Volume and Issue: 20(5), P. e0321257 - e0321257

Published: May 20, 2025

Efficient path planning and obstacle avoidance in a complex dynamic environment is one of the key challenges unmanned vehicle logistics distribution, especially scene Cainiao Station, which involves crowded communities campus roads. In view shortcomings existing methods multi-sensor data fusion optimization, this paper proposes model based on image fusion, named DynaFusion-Plan. The able to provide an optimal from starting point target environment, avoiding obstacles realizing smoothness adjustment ability path. consists three modules: sensor module uses Convolutional Neural Networks (CNN) Lidar-Inertial Odometry Simultaneous Localization Mapping (LIO-SAM) technology build high-precision map; combines Artificial Potential Field (APF) Deep Deterministic Policy Gradient (DDPG) algorithms balance length, smoothness, capabilities; decision control Model Predictive Control (MPC) Long Short-Term Memory (LSTM) achieve real-time tracking adjustment. Experimental results TartanAir, NuScenes, AirSim datasets show that DynaFusion-Plan significantly outperforms indicators such as length (42.5 m vs. 48.7 m), ( κ=0.05 id="M2">κ=0.15 ), success rate (98.7% 85.4%), environments. It shows strong adaptability stability. This work provides efficient reliable solution for intelligent scenarios, lays foundation future optimization directions, lightweight design more real-world scenario verification.

Language: Английский

Citations

0
CoroYOLO: a novel colorectal cancer detection method based on the Mamba framework DOI Creative Commons
Wenfei Chen,

Fengrui Hou,

Yue Shen

et al.

Frontiers in Physics, Journal Year: 2025, Volume and Issue: 13

Published: May 22, 2025

Colorectal cancer (CRC) is one of the most common malignant tumors worldwide, and early detection crucial for improving cure rates. In recent years, object methods based on convolutional neural networks (CNNs) transformers have made significant progress in medical image analysis. However, CNNs limitations capturing global contextual information, while can handle long-range dependencies, their high computational complexity limits efficiency practical applications. To address these issues, this paper proposes a novel model—CoroYOLO. CoroYOLO builds upon YOLOv10 architecture by incorporating concept State Space Model (SSM) introduces TSMamblock module, which dynamically models input data, reduces redundant computations, improves both accuracy. Additionally, integrates Efficient Multi-Scale Attention (EMA) mechanism, adaptively strengthens focus critical regions, enhancing model’s robustness complex images. Experimental results show that after training SUN Polyp PICCOLO datasets, outperforms existing mainstream Etis-Larib dataset, achieving state-of-the-art performance demonstrating effectiveness colorectal detection.

Language: Английский

Citations

0