Abstract
Crack segmentation is critical for maintaining structural safety, but fully supervised deep learning approaches require extensive pixel-level annotations, making large-scale deployment costly. This study adapts semisupervised learning framework for crack segmentation that extends weak-to-strong consistency regularization with domain-aware hard negative sample (HNS) integration to minimize annotation needs while improving robustness. The framework utilizes a dual-stream architecture with weak-to-strong image- and feature-level perturbation-based consistency regularization, adapted from general semantic segmentation to the domain-specific challenges of structural crack inspection. HNS, which are visually deceptive non-crack patterns such as shadows, stains, and surface textures, were integrated into the unlabeled training stream, enabling the model to better distinguish cracks from background noise and substantially reduce false positives. The proposed method achieved performance comparable to that of a fully supervised model using only 20% of the labeled data, effectively reducing the labeling costs by 80%. Experiments on concrete and asphalt structures demonstrate that HNS integration improves precision, especially in visually complex conditions, while a mixed-domain model trained on both surfaces achieves strong generalization. Furthermore, the proposed method was compared with other state-of-the-art semi-supervised methods and consistently achieved improved performance in both concrete and asphalt datasets. These findings indicate that integrating SSL with HNS enhances crack segmentation, providing a scalable solution for real-world applications where background complexity remains challenging.
Key Words
concrete and asphalt structures; consistency regularization; crack segmentation; hard negative samples (HNS); semi-supervised learning (SSL)
Address
(1) Muhammad Tanveer:
Department of Civil Engineering, University of Seoul, Dongdaemun-gu, Seoul 02504, South Korea;
(2) Soojin Cho:
Graduate School of Urban Bigdata Convergence, University of Seoul, Dongdaemun-gu, Seoul 02504, South Korea.
Abstract
To address the problems of low accuracy and weak anti-interference ability in bridge damage identification under complex noise environments, a damage identification method combining EMD-wavelet threshold joint denoising with curvature mode difference is proposed. First, empirical mode decomposition (EMD) adaptively separates high-frequency noise components from useful features in vibration signals. Subsequently, precise denoising is achieved using 3-level decomposition with db4 wavelet basis function and soft threshold processing. Finally, damage localization and quantitative prediction of damage severity are accomplished based on the curvature mode difference index. A finite element model is established using a 147 m-span variable-section continuous beam bridge as the prototype, and four types of working conditions involving single/double damage locations and different damage severities are designed. Numerical simulation verification is conducted under 5% intensity Gaussian white noise interference, with comparisons made against the single wavelet threshold denoising method and the EMD-SVD (Singular Value Decomposition) joint method. Results show that the single wavelet threshold method fails in multiple working conditions under noisy environments, while the proposed joint method effectively suppresses noise interference. It achieves 100% damage localization accuracy and controls the damage severity identification error within 3%~8.7%, improving accuracy by over 50% compared to the single wavelet threshold method. Additionally, it exhibits good adaptability to the complex structure of variable-section continuous beam bridges without requiring complex parameter adjustments, providing an efficient and reliable technical solution for bridge structural health monitoring.
Abstract
The primary structural element used by fingerprint recognition systems is the ridge orientation structure which provides a geometric prior for enhancing images, extracting features and comparing them to one another. However, most of the previous adversarial studies have operated in the pixel domain and therefore, structural manipulation within the orientation manifold has been left mostly unexplored. Therefore, we introduce a ridge orientation perturbation framework that is constrained based on ridge flow characteristics such that the generated perturbations preserve the smoothness and singularity properties of the original ridge flows. In doing so, we ensure that our attacks are biometrically plausible but induce instability to the verification process. We tested the proposed attack on controlled (FVC2004) and forensic latent (NIST SD27) databases using both classical minutiae-based and CNNbased matchers. Our experimental results show that severe degradation can be achieved when structurally consistent perturbations are applied, where the equal error rate (EER) increased on the NIST SD27 database. Our findings indicate that ridge orientations represent a critical structural component of fingerprint systems that is shared across multiple recognition paradigms and illustrate a previously under-characterized geometric vulnerability.
Address
(1) Enshirah Altarawneh:
Department of Computer Engineering, Faculty of Engineering, The Hashemite University, Zarqa, Jordan;
(2) Jawdat S. Alkasassbeh, Khalaf Y. Alzyoud:
Department of Electrical Engineering, Faculty of Engineering Technology, Al-Balqa Applied University, Amman, Jordan;
(3) Sattam Almatarneh:
Faculty of Information Technology, Zarqa University, Zarqa, Jordan;
(4) Redhwan Algabri:
Department of Computer Science and Engineering, Sejong University, Seoul, 05006, Republic of Korea.
Abstract
This study presents a comparative development of single- and multi-output Artificial Neural Network (ANN) surrogate models for accurately predicting the first five natural frequencies of a rotating Timoshenko beam. Specifically, a comprehensive dataset consisting of 17,576 samples was generated using a high-fidelity Chebyshev spectral collocation method over a wide range of dimensionless design parameters. An initial correlation analysis has been conducted to effectively investigate the statistical relationships between the design parameters and the associated first five natural frequencies. Based on the initial screening, two ANN modeling strategies have been established: a unified multi-output ANN model capable of simultaneously predicting all five natural frequencies, and a set of five independent single-output ANN models, each optimized for a specific natural frequency. The models have been trained, validated, and evaluated using a 70-15-15 data split and compared against standard performance metrics from the literature, including the Root Mean Square Error (RMSE), Mean Absolute Percentage Error (MAPE), and Coefficient of Determination (R2), along with the computational efficiency required for the model development. The results demonstrate that both ANN approaches significantly outperform a baseline linear regression model, confirming the strongly nonlinear nature of the problem under study. The single-output ANN models consistently achieved very high prediction accuracy across all modes, with R
Address
(1) Sameer Al-Dahidi, Ma'en S. Sari:
Department of Mechanical and Maintenance Engineering, School of Applied Technical Sciences,
German Jordanian University, Amman 11180, Jordan;
(2) Mohammad Alrbai:
Department of Mechanical Engineering, School of Engineering, University of Jordan, Amman 11942, Jordan.
Abstract
Energy harvesters are a promising solution for powering low-frequency, low-power autonomous devices. However, their nonlinear electromechanical behavior makes accurate modeling, performance prediction, and rapid design optimization challenging tasks, especially using conventional analytical and numerical techniques. These techniques are often constrained by simplifying assumptions and high computational cost. This study proposes a Deep Neural Network (DNN)–based and a Feedforward Neural Network (FNN)-based modeling framework for fast and accurate prediction of the dynamic response and harvested power of electret-based microcantilever energy harvesters under a wide range of operating conditions without reliance on simplifying assumptions. A comprehensive dataset comprising 27 high-fidelity numerical simulations of a single-degree-of-freedom model is employed for a wide range of operating conditions, including variations in excitation frequency, electret surface voltage, base acceleration, and electrical resistance load. The proposed DNN and FNN models are benchmarked against a numerical integration technique. Results demonstrate that the DNN accurately captures key nonlinear phenomena, including resonance shifts, bandwidth variation, nonlinear softening effects, and pull-in instability, while providing near-instantaneous predictions of performance once trained. Quantitatively, the DNN achieves a total Root Mean Square Error (RMSE) and a total Mean Absolute Error (MAE) of 1.10 × 10−5 and 3.73 × 10−6, respectively, on training data, outperforming the FNN by more than one order of magnitude. Furthermore, the DNN exhibits superior generalization to unseen operating conditions and maintains robust predictive capability in nonlinear regimes. By eliminating the need for simplifying assumptions and computationally intensive simulations, the proposed approach offers a powerful and efficient alternative for rapid design optimization, real-time implementation, and adaptive control of electret-based microcantilever energy harvesting systems. This work addresses a critical gap in the literature and underscores the potential of deep learning for advancing intelligent, self-powered microsystems.
Key Words
Artificial Intelligence; Deep Neural Network; electret; energy harvesters; Feedforward Neural Network; microcantilever; numerical integration
Address
(1) Bashar Hammad:
Department of Mechanical and Maintenance Engineering, German Jordanian University, Amman, 11182, Jordan;
(2) Ahmad Jobran Al-Mahasneh:
Department of Mechatronics Engineering, Philadelphia University, Amman, 19392, Jordan;
(3) Mohammed Abu Mallouh:
Department of Mechatronics Engineering, Faculty of Engineering, The Hashemite University, Zarqa, Jordan;
(4) Basel Jouda:
Department of Electrical Engineering, University of Prince Mugrin, Madinah, 42241, Saudi Arabia.
open access
