Advances in computer AI-assisted multimodal data fusion techniques DOI Creative Commons
Fan Pan, Qiang Wu

Applied Mathematics and Nonlinear Sciences, Год журнала: 2024, Номер 9(1)

Опубликована: Янв. 1, 2024

Abstract Through the integration of multimodal data fusion technology and computer AI technology, people’s needs for intelligent life can be better met. This paper introduces alignment perception algorithm fusion, which is based on combining model. Taking air pollutant concentration prediction as an example, time series obtained through LSTM model prediction, attention mechanism introduced to establish numerical pollution. Different stations are also selected acquire weather image data, TS-Conv-LSTM spatio-temporal quality images constructed by utilizing Conv-LSTM cell encoder, then TransConv-LSTM cell, integrates anti-convolution long-short-term memory network a decoder. The Gaussian regression was used combine models, thus achieving synergistic concentrations. RMSE ATT-LSTM dataset reduced 8.03 compared comparison model, predictive fit above 0.75 all R² values. lowest MAE value collaborative only 3.815, highest up 0.985. Introducing deep learning techniques into helps explore massive more deeply obtain comprehensive reliable information about it.

Язык: Английский

YOLOv5s-Based Lightweight Object Recognition with Deep and Shallow Feature Fusion DOI Open Access
Guili Wang, Chang Liu, Lin Xu

и другие.

Electronics, Год журнала: 2025, Номер 14(5), С. 971 - 971

Опубликована: Фев. 28, 2025

In object detection, targets in adverse and complex scenes often have limited information pose challenges for feature extraction. To address this, we designed a lightweight extraction network based on the Convolutional Block Attention Module (CBAM) multi-scale fusion. Within YOLOv5s backbone, construct deep maps, integrate CBAM, fuse high-resolution shallow features with features. We also add new output heads distinct structures classification localization, significantly enhancing detection performance, especially under strong light, nighttime, rainy conditions. Experimental results show superior performance scenes, particularly pedestrian crossing weather low-light Using an open-source dataset from Shanghai Jiao Tong University, our algorithm improves crossing-detection precision (AP0.5:0.95) by 5.9%, reaching 82.3%, while maintaining speed of 44.8 FPS, meeting real-time requirements. The source code is available at GitHub.

Язык: Английский

Процитировано

1
Generative AI-Based Real-Time Face Aging Simulation for Biometric Systems DOI Creative Commons

R J Anandhi,

Alok Pal Jain,

B. Ravali Reddy

и другие.

E3S Web of Conferences, Год журнала: 2025, Номер 619, С. 03004 - 03004

Опубликована: Янв. 1, 2025

Facial recognition is, therefore, a crucial aspect of biometric systems used when authenticating as well verifying people’s identity. But here natural aging increases number difficulties concerning accuracy and long-term reliability the control stated above. In this paper, new method real-time face simulation in context variance using Generative AI; specifically, GANs, is proposed. The proposed model tries to use generative AI generation improved synthetics with modified age appearance, allowing capture or antiaging changes facial features. This approach assessed experimentally from one database another datasets principal area interest future faces long run respect groups. work also looks at strength robustness for problems. outcomes presented show that applying AI-based system paradigm improves performance specifically addressing variations thus proposing valuable solution age- related paper considers some possible consequences security, privacy, concerns practical application real systems.

Язык: Английский

Процитировано

0
Optimized Fingerprint Crime Detection using Robust Deformed Convolutional Neural Network for 5G Network Secure Smart Cities DOI

Krishnakumar K,

Suli Ma,

Gritzalis Dimitris A.

и другие.

Knowledge-Based Systems, Год журнала: 2025, Номер unknown, С. 113342 - 113342

Опубликована: Март 1, 2025

Язык: Английский

Процитировано

0
A Secure Multi‐Model Biometrics Using Deep Learning Model Based‐Optimal Hybrid Pattern by the Heuristic Approach DOI
J Samatha, Madhavi Gudavalli

Transactions on Emerging Telecommunications Technologies, Год журнала: 2025, Номер 36(4)

Опубликована: Апрель 1, 2025

ABSTRACT A new Deep Learning (DL)‐based privacy preservation method using multimodal biometrics is implemented in this work. Here, the fingerprint, iris, and face are aggregated initial phase fed to Optimal Hybrid Pattern, where Local Gradient Pattern Weber used. Thus, two sets of patterns from diverse techniques for face, iris attained. Fitness‐aided Random Number Cheetah Optimizer (FRNCO) used optimization also selecting optimal Pixels attain pattern. Further, these three pattern images histogram‐based features, same FRNCO model optimization. It then forwarded final Bayesian Network (DBN) with a Gated Recurrent Unit (GRU) termed DB‐GRU approach acquiring classified outcomes. The designed assimilated recognize efficacy developed model.

Язык: Английский

Процитировано

0
FT‐HT: A Fine‐Tuned VGG16‐Based and Hashing Framework for Secure Multimodal Biometric System DOI
Seema Rani, Neeraj Mohan, Priyanka Kaushal

и другие.

Transactions on Emerging Telecommunications Technologies, Год журнала: 2025, Номер 36(5)

Опубликована: Апрель 24, 2025

ABSTRACT Multimodal biometric systems offer several advantages over unimodal systems, including a lower error rate, greater accuracy and broader coverage of residents. However, the multimodal need to store multiple traits associated with each user, which brings higher for integrity privacy. This study describes deep learning (DL) model feature‐level coalition that utilizes biographical data user's face iris create secure template. To reliable, unique shareable latent image, hashing (linearization) approach is used fusion architecture. Furthermore, hybrid architecture fuses sketching techniques erasable features integrates them into complete security framework in this work. The efficiency recommended method demonstrated using images from database. proposed provides ability delete templates better protect data. works “WVU” “hashing” “image retrieval.” improved VGG16 achieves 99.85. paper also information on structuring modalities such as hashing, techniques. further studies are needed extend other unrestricted aspects.

Язык: Английский

Процитировано

0
Multilevel parallel attention knowledge distillation for multimodal biometric recognition DOI
Lu Kong, Ruizhi Wu, Ergude Bao

и другие.

Engineering Applications of Artificial Intelligence, Год журнала: 2025, Номер 154, С. 110865 - 110865

Опубликована: Апрель 29, 2025

Процитировано

0
Unifying Heartbeats and Vocal Waves: An Approach to Multimodal Biometric Identification At the Score Level DOI
Hatem Zehir, Toufik Hafs, Sara Daas

и другие.

Arabian Journal for Science and Engineering, Год журнала: 2025, Номер unknown

Опубликована: Май 21, 2025

Язык: Английский

Процитировано

0
Optimization of 2D and 3D facial recognition through the fusion of CBAM AlexNet and ResNeXt models DOI
Imen Labiadh, Larbi Boubchir, Hassene Seddik

и другие.

The Visual Computer, Год журнала: 2024, Номер unknown

Опубликована: Дек. 6, 2024

Язык: Английский

Процитировано

1
Advances in computer AI-assisted multimodal data fusion techniques DOI Creative Commons
Fan Pan, Qiang Wu

Applied Mathematics and Nonlinear Sciences, Год журнала: 2024, Номер 9(1)

Опубликована: Янв. 1, 2024

Abstract Through the integration of multimodal data fusion technology and computer AI technology, people’s needs for intelligent life can be better met. This paper introduces alignment perception algorithm fusion, which is based on combining model. Taking air pollutant concentration prediction as an example, time series obtained through LSTM model prediction, attention mechanism introduced to establish numerical pollution. Different stations are also selected acquire weather image data, TS-Conv-LSTM spatio-temporal quality images constructed by utilizing Conv-LSTM cell encoder, then TransConv-LSTM cell, integrates anti-convolution long-short-term memory network a decoder. The Gaussian regression was used combine models, thus achieving synergistic concentrations. RMSE ATT-LSTM dataset reduced 8.03 compared comparison model, predictive fit above 0.75 all R² values. lowest MAE value collaborative only 3.815, highest up 0.985. Introducing deep learning techniques into helps explore massive more deeply obtain comprehensive reliable information about it.

Язык: Английский

Процитировано

0