Linda Mohaisen

The rapid growth of Internet of Things (IoT) deployment has increased exposure to Distributed Denial of Service (DDoS) attacks, while the latency and bandwidth costs of cloud-centric intrusion detection constrain real-time response. This paper presents a fog-based DDoS detection framework that combines XGBoost with information-gain feature selection. The framework uses 23 of the 78 features, a 70.5% reduction, at a measured accuracy cost of 0.007 percentage points. The evaluation used a sample drawn from every file of the CICDDoS2019 release, with benign traffic retained and attack traffic sampled to an attack-to-benign ratio of approximately 10:1. Feature selection, hyperparameter tuning, and threshold calibration were confined to the training partition. Under a random stratified split, the model reached 99.98% accuracy at a 0.058% false-positive rate. Under the capture-day split designated by the dataset authors, accuracy fell to 91.61%, and the false-positive rate rose to 8.82%, equivalent to 8820 false alerts per 100,000 benign flows. Per-request inference on a two-core profile took 1.0 to 5.2 ms over three repeated runs; feature extraction and network transit were not measured. Deployment would require monitoring for distribution shift and periodic retraining.

Most UAV missions currently require visiting multiple checkpoints to perform field tasks in environments with varying levels of obstacle complexity. These missions become more challenging because UAVs have limited onboard resources, particularly in terms of energy, making it necessary to determine a safe and efficient path that enables all required visits to be completed while minimizing both travel distance and energy consumption. To address these challenges, this study proposes a hybrid fuzzy metaheuristic approach that integrates Cuckoo Search and Tabu Search for multi-objective UAV path planning. The proposed approach generates collision-free paths in environments with static obstacles and employs fuzzy logic to construct a unified evaluation function, in which distance and energy values are mapped to membership functions and combined into a single fitness score to guide the optimization process. Cuckoo Search drives global exploration of the solution space, while Tabu Search refines solutions locally. Together, they improve path quality and avoid premature convergence. Experimental results across two scenarios with varying obstacle densities and checkpoint counts demonstrate the efficacy of the proposed hybrid approach. Compared with two baseline algorithms, the hybrid approach achieves reductions in path length ranging from 0.01% to 42.11% and in energy consumption ranging from 0.08% to 27.91%, depending on scenario complexity. Moreover, it maintains a high success rate of 96–100% as both checkpoint counts and obstacle density increase, whereas the baseline algorithms drop to 3–13% in more complex environments. These results highlight the effectiveness and scalability of the approach for multi-checkpoint UAV path planning in obstacle-rich environments.

Face masking, face swapping, and face animation are downstream activities that can benefit from face parsing, in which a face image is divided into many semantic regions. Due to the widespread use of cameras, obtaining facial images has become increasingly straightforward. However, pixel-by-pixel manual labeling takes a lot of time and effort, that allows us to investigate approaches based on unlabled data. In this paper, we propose a novel hybrid transfer learning-based approach for face parsing. First, patches are randomly masked in the central region of the face images. The method then proceeds in two stages: a pre-training stage and a fine-tuning stage. In the pre-training stage, the model is able to represent some basic facial features through unlabeled data. Then, the model is adjusted for the face parsing task on a small labeled dataset in the fine-tuning stage. Experimental results on the test sets show that the model can significantly reduce labeling costs. Furthermore, the proposed method outperforms the baseline by 2.9%, 2.16%, and 1.18% of mIoU with 0.5%, 1%, and 10% labeled data, respectively, on the LaPa dataset. Moreover, experimental results on the CelebAMask-HQ test dataset reveal that the masked transfer learning-based approach significantly outperforms the baseline for various labeling samples of the training data.

Hyperspectral imaging is a valuable technique for accurately classifying materials because of the abundance of spectral information and high resolution it provides. However, the characteristics of Hyperspectral Imaging, such as high-dimensional features and information redundancy, pose significant challenges to data processing. Traditional dimensionality reduction methods often have information loss, high computational complexity, and easy to ignore the strong correlation between HSI bands when dealing with the HSI data. Although other methods can achieve satisfactory classification performance, they do not consider the dimensionality reduction of HSI, and they focus on the model performance, which limits further improvement in classification performance. This article proposes a transformer-based framework called “SpectrumRecombineFormer” (SRF), which is composed of two key modules, namely “Spatial–Spectral Recombination” (SSRC) and “Cross-Layer Fusion” (CF). The SSRC is capable of utilizing both adjacent and non-adjacent spectrums to generate the spatial-sequential perceptive representations, which alleviate the effect of the strong correlation between HSI bands. The CF can avoid the loss of information during the feed-forward procedure among layers. Extensive experiments on five existing datasets (widely adopted Indian Pines, Houston2013, Pavia University, Salinas, and KSC) demonstrate the capability of our proposed method to address the above-mentioned challenges. Both quantitative and qualitative experimental ablation studies, including visualization results, reveal that the proposed SRF method can successfully and efficiently classify HSIs and surpass the other state-of-the-art methods. For access to the source code, please visit https://github.com/kangpeilun/SRF-HSI-Classification-master .

The Internet of Things (IoT) aims to wirelessly connect billions of physical things to the IT infrastructure. Although there are several radio access technologies available, few of them meet the needs of Internet of Things applications, such as long range, low cost, and low energy consumption. The low data rate of low-power wide-area network (LPWAN) technologies, particularly SigFox, makes them appropriate for Internet of Things applications since the longer the radio link’s useable distance, the lower the data rate. Network reliability is the primary goal of SigFox technology, which aims to deliver data messages successfully through redundancy. This raises concerns about SigFox’s scalability and leads to one of its flaws, namely the high collision rate. In this paper, the goal is to prevent collisions by switching to time division multiple access (TDMA) from SigFox’s Aloha-based medium access protocol, utilizing only orthogonal channels, and eliminating redundancy. Consequently, during a designated time slot, each node transmits a single copy of the data message over a particular orthogonal channel. To achieve this, a multi-agent, off-policy reinforcement learning (RL) Q-Learning technique will be used on top of SigFox. In other words, the objective is to increase SigFox’s scalability through the use of Reinforcement Learning based time slot allocation (RL-SCAP). The findings show that, especially in situations with high node densities or constrained communication slots, the proposed protocol performs better than the basic SCAP (Slot and Channel Allocation Protocol) by obtaining a higher Packet Delivery Ratio (PDR) in average of 60.58%, greater throughput in average of 60.90%, and a notable decrease in collisions up to 79.37%.

Locating available on-street parking in busy urban areas is often frustrating and time-consuming, particularly during peak hours. Inefficient parking searches contribute to traffic congestion, increased fuel consumption, and higher carbon emissions, making real-time parking availability information essential for smart urban mobility. This paper presents FedPark, a privacy-preserving crowdsensing system that leverages drivers’ smartphone sensors to monitor parking space occupancy. FedPark employs a Federated Learning (FL) algorithm, enabling the creation of a global model that aggregates insights from multiple local models maintained on individual drivers’ devices. Each local model learns from a user’s driving patterns without sharing personal data, ensuring enhanced privacy protection. Instead of transmitting raw data, only model updates are shared, reducing communication overhead and security risks. FedPark is evaluated under different urban conditions such as different numbers of drivers and destinations. The evaluation demonstrates that FedPark achieves a high level of accuracy at 98.31%, recall of 98.31%, precision of 98.34%, and an F1-score of 98.31% comparable to centralized machine learning models, including XGBoost, Random Forest, CatBoost, CNN, and ensemble classifiers while preserving driver privacy.

Changing a person’s posture and low resolution are the key challenges for person re-identification (ReID) in various deep learning applications. In this paper, we introduce an innovative architecture using a dual attention network that includes an attention module and a joint measurement module of spatial-temporal information. The proposed approach can be classified into two main tasks. Firstly, the spatial attention feature map is formed by aggregating features in the spatial dimension. Additionally, the same operation is carried out on the channel dimension to form channel attention feature maps. Therefore, the receptive field size is adjusted adaptively to mitigate the changing person posture issue. Secondly, we use a joint measurement method for the spatial-temporal information to fully harness the data, and it can also naturally integrate the information into the visual features of supervised ReID and hence overcome the low resolution problem. The experimental results indicate that our proposed algorithm markedly improves the accuracy in addressing changing human postures and low-resolution issues compared with contemporary leading techniques. The proposed method shows superior outcomes on widely recognized benchmarks, which are the Market-1501, MSMT17, and DukeMTMC-reID datasets. Furthermore, the proposed algorithm attains a Rank-1 accuracy of 97.4% and 94.9% mAP (mean Average Precision) on the Market-1501 dataset. Moreover, it achieves a 94.2% Rank-1 accuracy and 91.8% mAP on the DukeMTMC-reID dataset.