Article
Chemistry, Multidisciplinary
Michael Abebe Berwo, Yong Fang, Jabar Mahmood, Nan Yang, Zhijie Liu, Yimeng Li
Summary: This paper presents a robust approach using transfer learning and fine-tuning to classify cracks in automotive engine components. Two approaches, including building a Light ConvNet architecture from scratch and fine-tuning the top layers of ConvNet architectures, were investigated. Data augmentation was utilized to minimize over-fitting. The results showed that the transfer learning approach with fine-tuned MobileNet achieved better classification accuracy, and using YOLOv5s object detector with transfer learning effectively detected cracks in engine parts.
APPLIED SCIENCES-BASEL
(2022)
Article
Computer Science, Information Systems
Musarat Hussain, Chi Cheng, Rui Xu, Muhammad Afzal
Summary: Phishing scams are on the rise and require rapid, precise, and low-cost prevention measures. CNN-Fusion, a character-level convolutional neural network, is proposed as an effective and lightweight method for detecting phishing URLs. It utilizes parallel one-layer CNN variants with different-sized kernels and applies techniques like SpatialDropout1D and max-over time pooling to enhance its robustness and feature selection. Evaluation on publicly available datasets and against AI adversarial attacks shows superior performance compared to existing methods with significantly reduced training time and memory consumption, achieving an average accuracy above 99%.
INFORMATION SCIENCES
(2023)
Article
Computer Science, Artificial Intelligence
Taojiannan Yang, Sijie Zhu, Matias Mendieta, Pu Wang, Ravikumar Balakrishnan, Minwoo Lee, Tao Han, Mubarak Shah, Chen Chen
Summary: Existing deep neural networks are limited in their ability to perform inference at different resource constraints. In this work, the authors propose the MutualNet method, which trains a single network to run at various resource constraints. By training a cohort of model configurations with different widths and resolutions, MutualNet achieves consistent improvements on various tasks and datasets.
IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE
(2023)
Review
Computer Science, Information Systems
Syed Farooq Ali, Muhammad Aamir Khan, Ahmed Sohail Aslam
Summary: This study comprehensively reviews the fingerprint algorithms and techniques published in the last few decades, categorizing them into nine different approaches, with deep learning approach outperforming others. Based on 106 referred papers and their cumulative citations, fingerprint literature is historically divided into four eras.
FRONTIERS OF COMPUTER SCIENCE
(2021)
Article
Chemistry, Analytical
Sulabh Kumra, Shirin Joshi, Ferat Sahin
Summary: This study proposes a dual-module robotic system capable of generating and performing antipodal robotic grasps at real-time speeds. The model achieved state-of-the-art accuracy of 98.8%, 95.1%, and 97.4% on three standard datasets. Experimental results show significant improvement in grasp detection compared to prior work.
Article
Computer Science, Artificial Intelligence
Yue Niu, Annan Wang, Xuewu Wang, Shengxi Wu
Summary: Transformer-based networks dominate the field of pose estimation and outperform ConvNet-based networks. To revive ConvNets, ConvPose is proposed as a pure ConvNet that modernizes network structures instead of using attention mechanisms. The modernization process includes deepening layers, using separate convolution layers, applying batch normalization, using large-kernel separable convolutions, and designing re-parameterized-style structures. ConvPose achieves competitive results in terms of speed and performance compared to existing Transformer and ConvNet networks.
Article
Computer Science, Artificial Intelligence
Min Zhang, Haiyang Hu, Zhongjin Li, Jie Chen
Summary: This paper proposes an effective context-aware compositional ConvNet (CA-CompNet) for occluded workflow detection. By combining compositional model and original ConvNet together, utilizing bounding box annotations to segment the context from target workflow instance, and introducing a robust voting mechanism for candidate bounding box, the proposed method can robustly detect occluded workflow instances in the industrial environment.
Article
Computer Science, Theory & Methods
Xiongjun Guan, Yongjie Duan, Jianjiang Feng, Jie Zhou
Summary: Skin distortion is a challenge in fingerprint matching, and previous studies have focused on estimating and rectifying the distortion field to improve recognition rate. However, existing rectification methods based on principal component representation are not accurate and sensitive to finger pose. This paper proposes a rectification method using a self-reference based network to directly estimate the dense distortion field of distorted fingerprints, achieving state-of-the-art performance in distortion field estimation and rectified fingerprint matching.
IEEE TRANSACTIONS ON INFORMATION FORENSICS AND SECURITY
(2023)
Article
Computer Science, Artificial Intelligence
Gokhan Altan, Apdullah Yayik, Yakup Kutlu
Summary: Research focused on predicting EEG data using ConvNets and optimized with LuELM method, the results showed that using new machine learning methods can effectively improve prediction accuracy, outperforming traditional neural network models.
NEURAL PROCESSING LETTERS
(2021)
Article
Engineering, Electrical & Electronic
Yuyang Zhang, Yuangang Lu, Jianqin Peng, Chongjun He, Zelin Zhang
Summary: We propose a novel method to reduce the acquisition time of the Brillouin optical time-domain sensor (BOTDS) without increasing hardware complexity. The method uses sparse frequency sampling and an artificial neural network to accurately recover the Brillouin scattering spectrum (BSS) with low resolution. In a proof-of-concept experiment, the data acquisition time is reduced to 5.5% compared to normal sampling.
IEEE SENSORS JOURNAL
(2023)
Article
Computer Science, Artificial Intelligence
Lei Shi, Sheng Lan, Hao Gui, Yujiu Yang, Zhenhua Guo
Summary: Fingerprint recognition is an important biometric technology, with 2D contactless fingerprints receiving increasing attention due to their safety and hygiene advantages. However, low-resolution and posture inconsistency issues have hindered the performance of existing algorithms. A novel method named FTG has been proposed to address these challenges and achieve state-of-the-art performance.
Article
Chemistry, Multidisciplinary
Yuting Sun, Yanfeng Tang, Xiaojuan Chen
Summary: This paper proposes an attention-based partial fingerprint identification model APFI, which uses residual networks for feature extraction and inserts a channel attention module for more accurate fingerprint feature information. The similarity of fingerprints is calculated using the angular distance between features to improve identification accuracy. Experiments show that the proposed method outperforms other state-of-the-art methods in fingerprint identification accuracy.
APPLIED SCIENCES-BASEL
(2023)
Article
Computer Science, Information Systems
Da Li, Zhao Niu
Summary: Research on location technology, particularly signal-based fingerprint positioning, has gained significant interest due to its high performance. A hierarchical positioning system using wavelet feature images was proposed to enhance accuracy, with ResNet and MLP utilized for feature learning. Experimental results showed improved positioning performance outdoors through data enhancement techniques.
ISPRS INTERNATIONAL JOURNAL OF GEO-INFORMATION
(2021)
Article
Medicine, General & Internal
Ibtihal A. L. Alablani, Mohammed J. F. Alenazi
Summary: The COVID-19 pandemic has a significant impact on global health and well-being. Effective patient screening through chest radiography is crucial in combating the disease. This paper introduces COVID-ConvNet, a deep convolutional neural network designed to detect COVID-19 symptoms from chest X-ray scans. Experimental results show that the COVID-ConvNet model achieves a high prediction accuracy of 97.43%, outperforming recent related works by up to 5.9% in terms of prediction accuracy.
Article
Multidisciplinary Sciences
A. H. Abdul Hafez, Ammar Tello, Saed Alqaraleh
Summary: A new collaborative VPR approach is proposed in this study, which utilizes ROIs feature maps from two different layers to improve recognition performance. The results demonstrate the robustness of the proposed method compared to state-of-the-art methods in facing viewpoint and appearance challenges, with AUC and mAP measures achieving an average of 91%.
ARABIAN JOURNAL FOR SCIENCE AND ENGINEERING
(2022)