4 June 2025 in Knowledge-Based Systems
18 August 2025 in IEEE Access
Ensemble models for network anomaly detection are often challenged by class imbalance, a condition known to substantially degrade predictive accuracy. To address this limitation, a novel ensemble learning framework, termed the GMM Selector for Anomaly Detection (GMM-SAD), is proposed. The GMM-SAD framework utilizes Gaussian Mixture Models (GMM) to first estimate the prevalence of anomalies and then dynamically routes data to the most suitable specialized detection model within its architecture. To validate its effectiveness and generalization capabilities, extensive empirical evaluations were conducted on three widely-recognized benchmark datasets: CICIDS2017, UNSW-NB15, and NSL-KDD, under varying degrees of class imbalance. The proposed GMM-SAD model demonstrates robust and consistent performance, achieving an impressive average Area Under the Curve (AUC) of 98.40%, F1-score of 94.16%, and a high Recall of 95.32% across all datasets and imbalance scenarios. These findings confirm that the model significantly enhances network anomaly detection accuracy and robustness, particularly its ability to identify rare instances under challenging imbalanced conditions. This research introduces a highly adaptive technical approach for mitigating class imbalance, demonstrating strong potential for practical deployment in real-time network security monitoring.
5 June 2026 in Current Research in Food Science
Coffee quality assessment is critical to the global agricultural economy, yet traditional manual inspection remains labor-intensive, subjective, and prone to inconsistency. This study presents an improved YOLOv10-based framework specifically engineered for the automated detection of defective green coffee beans, achieving an optimal balance between high-precision localization and real-time edge deployment constraints. The proposed architecture integrates three targeted structural innovations: Depthwise Separable Convolution (DSConv) with distribution shift for lightweight, fine-grained feature extraction; a Spatial Pyramid Pooling Fast with Attention (SPPF_Attention) module for context-aware multi-scale feature refinement; and Partial Convolution (PConv) operators in the detection head to mitigate computational redundancy and enhance robustness against dense occlusions. Comprehensive evaluations utilized a rigorously curated dataset covering seven defect categories compliant with Specialty Coffee Association (SCA) standards. Evaluated under a 5-fold cross-validation protocol, the framework achieved state-of-the-art performance with 99.2% mAP@50, 98.5% precision, and 98.8% recall, alongside a rapid inference latency of merely 2.0 ms. These metrics denote substantial improvements over the advanced YOLOv11 baseline, simultaneously reducing model parameters by 21.6%, decreasing model size by 12.4%, and accelerating inference by 33.3%. Furthermore, systematic ablation studies validate the incremental contributions of each architectural component, while EigenCAM visualizations confirm that the integrated attention mechanisms successfully steer the network toward semantically meaningful defect regions and suppress background interference. Extensive cross-dataset analyses additionally demonstrate the model’s exceptional generalization capability across diverse processing states, establishing this framework as a highly viable, scalable solution for resource-constrained industrial sorting facilities.
11 July 2025 in IEEE Wireless Communications Letters
This letter investigates adaptive resource allocation in a covert Vehicle-to-Everything (V2X) communication system comprising a Vehicle-to-Vehicle (V2V) pair, multiple wardens, and multiple cellular vehicular users. The goal is to maximize total system capacity while satisfying covertness, power, bandwidth, and latency constraints. The optimal detection threshold and total detection error probability for the warden are first derived analytically. To solve the resulting constrained optimization problem, a decentralized Adaptive Constraint-Aware Multi-Agent Deep Deterministic Policy Gradient (ACA-MADDPG) framework is proposed. Each agent independently learns transmit strategies using local observations and constraint feedback, enabling scalable deployment in dynamic environments. Simulation results verify that the proposed method outperforms DQN-and DDPG-based benchmarks in both capacity and reward convergence.