Ad hoc networks are increasingly deployed in critical applications due to their flexibility and scalability. However, their decentralised and dynamic nature makes them highly vulnerable to a range of sophisticated security threats. This paper aims to improve the efficiency of intrusion detection and mitigation in ad hoc networks using an AI-driven deep learning approach. A hybrid deep learning model is proposed, integrating convolutional neural networks (CNNs) for feature extraction and long short-term memory networks (LSTMs) for temporal analysis to effectively detect malicious activities. Reinforcement learning, particularly using a deep Q-network (DQN), is applied to dynamically select optimal mitigation strategies. Federated learning is also used to train the model in a distributed manner, ensuring privacy while allowing scalability across network nodes. The proposed approach shows significant improvements in intrusion detection accuracy, exceeding 90 %, and offers effective real-time mitigation strategies. These results provide a comprehensive and adaptive framework for securing ad hoc networks against evolving threats.

Frequency estimation for a single complex sinusoid in noise is a fundamental problem in signal processing. A suboptimal but simple frequency estimator, known as Jacobsen estimator, which is based on three discrete Fourier transform (DFT) samples, gives good bias performance without the need to increase the DFT size. Candan has modified the Jacobsen estimator by adding a so-called bias correction factor to further reduce the bias of the estimator. In addition to bias considerations, a number of asymptotic variance expressions of the estimators were performed in the literature. However, these expressions are valid only for signal frequencies located very near a DFT bin index. In this paper, with the use of a simple variance analysis technique, an accurate general variance expression for arbitrary frequency locations is derived for the case of windowed data. A general method for calculating the bias correction factor is also proposed. The variance expression is examined for the cosine-sum window family. An approximate variance formula for sufficiently large data record lengths is also given for windows from this family. Computer simulations are included to validate the theoretical results.

In this work, two universal current-mode (CM) filters are proposed. These filters are single-input three-output filter and three-input single-output filter. The proposed single-input CM filter simultaneously provides low-pass filter (LPF), high-pass filter (HPF), and band-pass filter (BPF) responses without requiring passive component matching. The LPF and BPF provide gains. Additionally, a notch filter (NF) response can be obtained by subtracting the output current of the LPF from the output current of the HPF, while an all-pass filter (APF) response can be achieved by subtracting the output current of the BPF from the output current of the NF. The proposed three-input CM filter generates LPF, HPF, BPF, NF, and APF responses at its output when the inputs are appropriately selected. The BPF response of the three-input CM filter has a gain. However, HPF, NF, and APF responses require a passive component matching condition. Both proposed circuits contain two grounded capacitors, which makes them suitable for integrated circuit technology and exhibit low active and passive sensitivities. The performance of the proposed filters is verified through SPICE simulations using 0.18 µm Taiwan Semiconductor Manufacturing Company (TSMC) complementary metal-oxide-semiconductor (CMOS) technology parameters. Additionally, experimental validation is performed using AD844s to demonstrate their practical performance.

LiDAR point cloud-based place recognition (LPR) in unstructured natural environments remains an open challenge with limited existing research. To address the limitations of unstructured environments, such as sparse structural features, uneven point cloud density, and significant viewpoint variations, we present BSPR-Net, a dual-branch point cloud feature extraction approach for point cloud place recognition, which consists of a BEV - projection rotation - invariant convolution branch and a point cloud sparse convolution branch. This design enhances the representation capability of geometric structural features while aggregating rotation-invariant characteristics of point clouds, thereby better addressing the challenge of large viewpoint disparities in reverse-revisited unstructured environments. The proposed network was tested on multiple reverse-revisited sequences of the Wild-Places data set, a benchmark for unstructured natural environment place recognition. It achieved a maximum F1 score of 85.46 %, exceeding other classical methods by more than 4 %. The ablation experiments further confirmed the effectiveness of each module in improving place recognition performance.

Automatic detection of road cracks is essential for the long-term safety maintenance of roads, bridges, and other infrastructure. Although existing deep learning-based crack segmentation methods have improved detection accuracy, challenges remain in terms of high computational complexity and inadequate capture of fine cracks and edge details. To address these issues, this study proposes an enhanced UNet-based architecture, termed DynaEdge-Net. In the encoder and decoder stages, a Residual Detail Enhancement Block (RDEB) and a Cascaded Group Attention (CGA) module are incorporated to strengthen edge feature representation and focus on critical regions, respectively. In the skip connections, a Group-wise Dynamic Gating (GDG) module is introduced to adaptively suppress background noise and optimize feature transmission. During decoding, a Dynamic Upsampling (DySample) strategy replaces conventional interpolation, enabling high-fidelity reconstruction of crack structures. Experimental results show that DynaEdge-Net achieves IoU, F1-score, boundary F1-score, and recall rates of 82.34%, 90.83%, 83.27%, and 90.12%, respectively, outperforming several state-of-the-art road segmentation algorithms. The proposed method not only improves the continuity and accuracy of crack extraction but also demonstrates strong robustness and generalization capability, providing a reliable solution for intelligent inspection and maintenance of transportation infrastructure.

Transformers are critical and expensive components of power systems. Therefore, it is important that these systems operate at optimum performance levels and sustainable economic conditions. In the normal operating environments, mineral oil passes on a slow and natural deterioration, while under conditions of thermal or electrical stress, the deterioration ratio increases. Due to breakdown, the hydrocarbon gases H2, CH4, C2H6, C2H4, C2H2, CO, and CO2) are composed in the transformer mineral oil. There are several conventional methods for identifying and classifying incipient faults in power transformers based on dissolved gas analysis (DGA). However, these methods have the disadvantage of not being able to distinguish situations where multiple electrical or thermal faults occur simultaneously. Due to the serious disadvantage of traditional DGA methods in terms of accuracy and consistency estimation compared to algorithm calculations made with artificial intelligence techniques, researchers have started to work intensively on artificial intelligence techniques in recent years. This comprehensive review aims to combine and present in a single source basic information about classical methods of power transformer fault diagnosis for DGA, historical development of the devices used, artificial intelligence-based methods, accuracy classifications of predictions. This investigation also revealed the contribution of the parameter optimisation process to eliminate the imbalance of the dataset in the accuracy of prediction when applying artificial intelligence techniques in DGA. In this study, the prediction performance of each research method performed with artificial intelligence techniques in fault diagnosis among the compared methods was analysed. This review emphasises the importance of eliminating dataset imbalance by performing parameter optimisation in artificial intelligence technique with an in-depth research-orientated perspective. As a result, this study not only encourages new ideas, but also provides a comprehensive source of literature for future accessibility of the subject.

Microplastics and heavy metals are materials that harm the environment and living organisms. Rapid detection enables their control or the identification of their sources. Conventional detection methods are expensive and require expert interpretation. The proposed sensor system detects these materials and evaluates their concentrations using a trainable multilayer perceptron algorithm. The system consists of twenty-two different light spectrum LEDs and eighteen narrow bandwidth photodiodes. The absorbance of incoming light and the shifted bandwidths in the spectrum can be evaluated by assessing multiparametric optical events. The study examines water samples containing eight different heavy metals, namely arsenic (As), cadmium (Cd), chromium (Cr), copper (Cu), mercury (Hg), nickel (Ni), lead (Pb), and zinc (Zn), along with three types of microplastics: melamine particles with a diameter of 8 µm, polystyrene particles with a diameter of 8 µm, and polystyrene particles with a diameter of 10 µm. Classification was performed in a concentration-dependent and concentration-independent manner. The system performance was improved by selecting features using the Ranker method of the InfoGain algorithm. Measurements were performed without applying any indicator chemicals. The system demonstrated high success in concentration-independent evaluation and acceptable concentration accuracy for heavy metals.

Most existing methods of active tracking focus mainly on slow-moving objects, resulting in limited adaptability to objects with variable speeds. To bridge this research gap, a novel spherical coordinate guided adaptive active tracking (SCAAT) approach based on the pan–tilt–zoom (PTZ) camera machine vision system is proposed in this study. For object detection and tracking, YOLOv5 and DeepSORT are employed in the PTZ vision system. The spherical coordinates and angular speeds of the moving object can be acquired under the spherical coordinate system. For practical application, the start-time and start-angle delay of the PTZ cameras are calibrated, and a speed control equation is conducted in the spherical coordinate system to reduce rapid location deviation between the PTZ and the moving object. To adapt different speeds of the object and avoid camera shaking under different zooms, an adaptive tracking window is designed to keep the object within the camera field of view. Experimental testing has been performed to evaluate the proposed SCAAT method. The results indicate that the SCATT can not only expand the effective following distance and zoom of the PTZ camera, but also effectively improve the accuracy and stability of active tracking for the moving object with large speed variations.

Blockchain technology, fundamentally characterised by decentralisation, immutability, and consensus mechanisms, revolutionises data management and communication security. Unlike traditional centralised databases controlled by a single authority, blockchain distributes data across a network of nodes in a peer-to-peer framework, enhancing security by eliminating single points of failure. Immutability ensures that once data are recorded, they cannot be altered due to cryptographic hashing, where the hash of each block depends on its content, making tampering evident. Real-world applications include secure cargo tracking, identity management, and smart contracts. Blockchain enables real-time tracking of cargo, verified digital identities, and self-executing contracts that automate processes such as ownership verification and payment settlements. This article explores the potential of blockchain to fortify maritime communication security. Suggestion for a secure communication structure in shipping is presented, as well as a blockchain algorithm in Python for secure application onboard. For practical implementation, a blockchain-based document workflow management system can be developed to manage critical maritime documents.

Structural health monitoring (SHM) of offshore jacket platforms is crucial, and currently traditional deep learning methods such as artificial neural networks (ANNs) are widely used in damage identification of offshore conduit rack platform structures, which focusses on mapping feature information caused by damage to structural damage patterns. However, traditional methods have limitations in dealing with the time series data in the feature information. To improve the application of the time series information generated from offshore platform structures in damage modes, we propose a new integrated deep learning network model, which is used to improve the accuracy of the damage mode recognition based on the acceleration information of the conduit rack structure. First, the temporal convolutional network (TCN) breaks through the localisation of traditional convolutional neural networks in modelling the temporal dimension by efficiently extracting the long-term time since of the structural vibration response through an expansive causal convolution mechanism. Second, the bidirectional long short-term memory network (BiLSTM) further extracts the contextual information and global features of the data by extracting feature information in both directions and fusing the before and after correlations of vibration response signals. In addition, we adopt the Newton-Raphson-based optimiser (NRBO) optimisation algorithm for global optimisation of the hyperparameters of TCN and BiLSTM to avoid the subjectivity of manual parameter tuning, which significantly improves the model convergence speed and generalisation performance. Experimentally validated by finite element model simulation and testbed construction, our proposed NRBO-TCN-BiLSTM combined neural network damage identification accuracy is as high as 99 % on average, exceeding existing deep learning methods. The method has a wide range of applications in SHM for offshore platforms.