The deployment of deep neural networks on embedded systems remains a challenging task due to stringent constraints in computational resources, memory bandwidth, and energy efficiency. In this work, we present QYOLOv10, a quantization-aware acceleration framework designed to enable efficient object detection using low-precision variants of the YOLOv10 model-namely nano (n), small (s), and medium (m). The proposed method integrates advanced post-training quantization techniques for weights, activations, and inputs, with adaptive precision strategies aligned to hardware-specific constraints. These optimizations strike a favorable trade-off between computational efficiency and detection accuracy. To demonstrate real-world viability, we deploy QYOLOv10 on the Xilinx Kria KV260 platform using a customized Deep Processing Unit (DPU) overlay synthesized in Vivado and executed via the Vitis AI runtime. Experimental results show that QYOLOv10 achieves up to 4.8x improvement in inference speed, over 40% memory footprint reduction, and maintains accuracy within 1.2% of the original full-precision models. Inference latency is reduced to below 25 ms per image, enabling robust real-time performance. These results highlight the suitability of QYOLOv10 for edge-centric applications such as autonomous navigation, intelligent surveillance, and IoT-based visual analytics. By addressing the practical challenges of deep model deployment under resource-constrained conditions, this work contributes a scalable and hardware-aware solution for embedded object detection systems.

Because of the increased attack surfaces brought about by the proliferation of IoT devices, sophisticated intrusion detection systems (IDS) that can mitigate threats in real time are required. Although deep learning (DL) provides better attack pattern recognition, IoT environments with limited resources are challenged by its computational demands. The dual requirements of high classification accuracy and lightweight operation are difficult for conventional implementations to meet. This paper introduces a DL-based IDS architecture utilizing a Convolutional Neural Network (CNN), trained on the NF-BoT-IoT dataset’s realistic IoT traffic and deployed on a Raspberry Pi 4 model B using TensorFlow Lite (TFLite). The model is tested in both TFLite format and quantized INT8 to show the balance between accuracy, model size, and inference speed. Experimental results show that the TFLite and INT8 models outperform the original TFKeras model by reducing size by over 80%, speeding up inference by over 99%, and maintaining accuracy above 98.6%. The INT8 model achieves the optimal tradeoff between speed and compactness compared to TFlite. Thereby, it is the perfect choice for real-time intrusion detection in edge computing environments with limited resources. This research bridges the gap between powerful DL models and resource-constrained edge devices by developing a scalable, intelligent IDS suitable for real-world IoT deployments.

Recent years have seen a rapid development in Machine Learning, which has profoundly influenced many areas of science and engineering. Among them, computer vision takes the leading place, where important tasks are image classifications powered by CNNs. Despite the great performance of CNNs in complicated scenarios, they remain sensitive to so-called adversarial attacks, and deliberate perturbations leading them to incorrect predictions. Besides more innocuous consequences, this has serious security implications for critical applications, in-cluding medical diagnostics, where misclassifications might result in disastrous outcomes. This research work discusses adversarial attacks on CNNs and other DNNs in computer vision, studying a full range of the generation and detection methods with details while discussing intrinsic vulnerability and robustness. It also proposes a learning framework that will enhance the robustness and security of DNNs and CNNs against such adversarial perils. The ultimate goal is directed to an improvement in the reliability of such models in absolutely critical scenarios for safe deployment into applications where accuracy is crucial.

The widespread success of Convolutional Neural Networks (CNNs) in computer vision has been accompanied by soaring computational demands, often requiring high-performance GPUs for real-time inference. However, such hardware is impractical in embedded and resource-constrained environment. To address this, we propose a post-training quantization (PTQ) framework that converts CNN models from FP32 to INT8 without retraining, optimized for FPGA deployment. Using asymmetric quantization and TensorFlow Lite, we implemented VGG16 and ResNet50 on a PYNQ-Z1 Field-Programmable Gate Arrays (FPGA). The quantized VGG16 achieved a 67% increase in throughput (from 150 FPS to 250 FPS), a 68% reduction in latency, and a 52% improvement in Power-Delay Product. ResNet50 saw over 420% gain in DSP efficiency, a 3100% increase in LUT efficiency, and a 94% PDP reduction. Despite a marginal accuracy loss, both models showed significantly improved energy efficiency and performance-per-resource utilization. Our results confirm that PTQ enables scalable, low-power AI inference suitable for real-time applications on edge and embedded systems.

Over the last few years, the rapidary growth of the Internet of Things (IoT), has significantly increases the intricacy of managing network security. As IoT networks broaden, they become progressively vulnerable to cyberattacks, requiring advanced threat detection and mitigation solutions. To address these challenges, Intrusion Detection System (IDS) is fundamental for identifying and alleviating potential menaces in IoT environments. This paper, investigates the use of Deep Learning (DL) techniques, which have proven to be h effective in detecting and classifying malicious network traffic. Hence, we propose an IDS architecture that combines Convolutional Neural Networks (CNN) used to extract spatial features with four variants of Recurrent Neural Networks (RNN) including Long Short-Term Memory (LSTM), Bidirectional LSTM (BiLSTM), Gated Recurrent Unit (GRU), and Bidirectional GRU (BiGRU) utilized to capture temporal features. Together, these networks are able to anticipate and categorize malevolent cyberattacks in IoT network traffic more effectively than traditional methods. Experimental evaluation was conducted on the NF-UQ-NIDS, a comprehensive Netflow dataset that blends data from NF-BoT-IoT, NF-ToN-IoT, NF-UNSW-NB15, and NF-CSE-CIC-IDS2018 for binary classification. The efficacy of our proposed models was evaluated using various parameters such as accuracy, precision, recall, F1 score, False Alarm Rate (FAR) and Area Under Curve (AUC). A comprehensive comparative study was conducted that evaluated four proposed models in multiple datasets. The study underscores the potential of combining spatial and temporal DL models for advanced network security applications, demonstrating notable improvements in detection accuracy and reductions in false alarm rates. Consequently, the impressive results achieved by our models show the effectiveness of integrating CNN and different RNN variants for intrusion detection in IoT networks, offering a strong solution to protect IoT ecosystems against security threats.

With the swift progress of technology and the increasing frequency of cyber attacks on organizational networks and systems, cybersecurity has become one of the most critical challenges in the modern digital world. Within this framework, Intrusion Detection Systems (IDS) are critical to mitigate the impact of cybercrimes and safeguarding systems from various malicious attacks. However, traditional machine learning (ML) approaches have ceased to be sufficient to handle the complexities of large-scale and unstructured data, particularly in Internet of Things (IoT) environments. To bridge this gap, we propose a hybrid DL-based IDS that leverages the synergistic strengths of convolutional neural networks (CNNs) and gated recurrent units (GRU). While CNNs excel at extracting spatial features from network traffic data, GRUs effectively model temporal dependencies, making their combination particularly well-suited for detecting dynamic and complex IoT intrusions. In this study, we evaluate three DL models including CNN, GRU, and a hybrid CNN-GRU approach on two recent Netflow-based datasets, NF-UNSW-NB15 and NF-CSE-CIC-IDS2018 for binary classification. Extensive experiments using accuracy, precision, recall, F1-score, False Alarm Rate (FAR), AUC, and processing time metrics show that the CNN-GRU model outperforms standalone CNN and GRU models, as well as existing literature approaches, achieving an accuracy of 98.60% on the NF-UNSW-NB15 dataset and 97.95% on the NF-CSE-CIC-IDS2018 dataset. Our findings underscore the efficacy of combining spatial and temporal DL architectures for robust IoT intrusion detection, offering a scalable solution for modern cyber threats.

Recently, convolutional neural networks (CNNs) have shown remarkable performance in a variety of computer vision tasks. However, as CNNs become more complex, there is a higher demand for computational power, which typically necessitates the use of advanced hardware, limiting their scalability and broader applicability. This drives the need for optimization techniques to reduce these computational costs. We propose a uniform approach that enables the deployment of neural networks on FPGA platforms that cannot handle highprecision values. A common strategy to address this challenge is to perform low-precision computations through neural network quantization. The goal of this paper is to evaluate post-training integer quantization across different networks and criteria. We introduce an efficient framework designed to accelerate computations while simultaneously reducing both latency and memory overhead. This approach involves asymmetric quantization of the weight and activation matrices with the goal of minimizing these costs. Our experiments demonstrate a model size reduction of up to $75 \%$, with minimal accuracy degradation of less than $1 \%$ on benchmark datasets. Furthermore, we observe latency improvements of up to $3 \times$ compared to full-precision models. We believe that the proposed method will provide new insights into the interpretation of the quantization of neural networks.

Deep learning (DL) models have become prevalent and consistently exhibit outstanding performance across diverse domains, particularly in information security. As a subset of machine learning (ML), DL has proven adept at handling complex, unstructured, and large-scale data, making it well-suited to tackle the evolving challenges of cybersecurity. By leveraging DL techniques, we can develop robust intrusion detection systems (IDS) to counter a wide range of network attacks, serving as a crucial defense against hackers and cybercriminals. The integration of DL into IDS holds immense potential for significantly enhancing network security and mitigating cyber risks in various real-world scenarios. This work proposes an IDS architecture that integrates Convolutional Neural Network (CNN) for spatial feature extraction and Long Short-Term Memory (LSTM) for temporal feature extraction and sequential analysis to predict and classify malicious cyberattacks in IoT network traffic. Our experiments with the CSE-CIC-IDS2018 dataset, the latest comprehensive network traffic dataset, demonstrate the superiority of our approach in terms of accuracy, precision, recall, and F1 score. Our model achieves an unprecedented accuracy of over 99% in both binary and multiclass classification, surpassing existing efforts in the literature.

The complexity of data analytics applications has led to an exponential increase in demand for effective and high-performance data clustering solutions in recent years. This paper presents a technique for optimizing the speed of the widely-used K-means clustering algorithm, utilizing Field-Programmable Gate Arrays (FPGAs) and High-Level Synthesis (HLS). By taking advantage of FPGAs' parallel processing capabilities, a methodology built on the Open Computing Language (OpenCL) was presented, which significantly improved algorithmic efficiency and addressed the computational bottlenecks encountered in large-scale data clustering applications. We present a detailed overview of the OpenCL framework integration and the FPGA- based hardware acceleration framework for K-means algorithm optimization. Furthermore, this implementation can be fully realized without the need for Hardware Description Language (HDL) code. The performance improvements obtained using OpenCL-based FPGA acceleration are demonstrated through experimental evaluations on randomly generated float datasets, which demonstrate notable speedups over conventional software implementations. Experimental results demonstrate that FPGA- based accelerators offer significantly improved execution times, ranging from 3 to 16.7 times faster compared to traditional CPU implementations.

Due to the growing number of network devices, traffics and services, designing robust Intrusion Detection System (IDS) become a crucial need in the face of complex and various network attacks as a protective measure from hackers and cybercriminals. However, the traditional Machine Learning (ML) approach shows success in many research topics but with the increase in the amount of data and the diversification of network threats methods, it seems to lack reliability and accuracy. Therefore, considering a large amount of real-world cyber traffic, Deep Learning (DL) may be able to extract big data features more effectively. In this paper, we suggest an intrusion detection system for the Internet of Things (IoT) network based on Deep Learning to recognize various assault types for both binary and multiclass classification using two variants of Recurrent Neural Network (RNN) models long short-term memory (LSTM) and Bidirectional LSTM (BiLSTM). We have experimented the models with CSE-CIC-IDS2018, which is the newest comprehensive network traffic dataset. Accuracy, precision, recall, and F1 score are a few performance criteria where the suggested approach clearly excels. After comparison, we can infer that Bi-directional LSTM outperforms LSTM and other existing efforts in the literature. The accuracy of the experimental results was high, coming in at 98.62%.

Great interest is directed toward real-time applications to determine the exact location of sensor nodes deployed in an area of interest. In this paper, we present a novel approach using a combination of the Kalman filter and regularized bounding box method for localizing unknown nodes in an area using an FPGA-enhanced edge computing UAV whose trajectory is known and is represented as the position of many anchors. The UAV is equipped with a GPS system that allows it to gather location data of sensor nodes as it moves around its environment. We employ a regularized bounding box to predict the positions of the unknown nodes using regularization factors and we use the Kalman filter algorithm to smooth and improve the accuracy of the sensor nodes to be localized. In order to localize the unknown nodes, the UAV receives the number of hops from each node and uses this information as input to the localization algorithm. Furthermore, the use of an FPGA board allows for real-time processing of sensory data, enabling the UAV to make fast and accurate decisions in dynamic environments. The localization algorithm was implemented on the FPGA board “Zynq MiniZed 7007s evaluation board” using Xilinx blocks in Simulink, and the generated code was converted into VHDL using Xilinx System Generator. The algorithm was simulated and synthesized using “Vivado” software. In fact, the proposed system was evaluated by comparing the performances achieved through two different implementations: Hardware and Software implementation. In effect, the performance of FPGA hardware implementation presents a new achievement in localization due to its easy testing and fast implementation. Our results show that this approach can efficiently locate unknown nodes with good latency and high accuracy. In fact, the execution time of the FPGA-integrated algorithm is reduced by about 60 times compared to the software implementation and the power consumption is about 100 mW, which proves the suitability of FPGA for localization in WSNs, offering a promising solution for various mobile WSN applications.

Localization is a crucial concern in many wireless sensor networks (WSN) applications. Moreover, getting accurate information about the geographic positions of nodes in the sensing field is necessary in order to make the collected data useful and meaningful. In this context, the heuristic distance vector called DV-Distance is widely used for range-free localization in multi-hop WSN thanks to its advantages such as its simplicity and good accuracy. However, its localization accuracy may be relatively low due to environmental conditions. In this paper, we propose an improvement of the DV-Hop algorithm based on distance computation via a received signal strength indicator (RSSI) and a simple polynomial approximation technique. The proposed technique consists of RSSI measurement for distance estimation between sensor nodes thanks to low cost and simplicity compared to other techniques based on time and angle of arrival, i.e. TOA, AOA and TDOA which require more specialized equipment. However, as RSSI values may be affected by conditions environment such as noise, we propose to use the Kalman Filter in order to deal with this issue and improve localization accuracy. Test results proved that the proposed technique has a higher localization accuracy than DV-Distance by up to 50% and by up to 18% when compared with the state-of-the-art algorithms cited above and other DV-Hop-based localization methods.

As localization represents the main core of various wireless sensor network applications, several localization algorithms have been suggested in wireless sensor network research. In this article, we put forward an iterative bounding box algorithm enhanced by a Kalman filter to refine the unknown node’s estimated position. In fact, several research efforts are currently in progress to extend the 2D positioning algorithm in WSNs to 3D that reflects reality and the most practical applications. Subsequently, we replace a large number of GPS-equipped anchors with a single mobile anchor. In our studies, we consider the type of range-free sensor network exploiting the wireless sensors connectivity. We assess the performance of our algorithm using exhaustive experiments on several isotropic and anisotropic topologies. Our proposed algorithm can fulfill the joint goals of algorithm transparency and accuracy for various scenarios by evaluating parameters such as localization accuracy whilst changing other simulation parameters such as the effect of communication range, mobile anchor node position and sensor node deployment topology. It has been proven by the results of the experiments that the proposed algorithm effectively reduces the location error without requiring more equipment or increasing the communication cost.

The Internet of Things (IoT) has developed rapidly and been integrated with a variety of domains. Such a technology allows devices to send, receive, and process data without human involvement. Even though IoT has been widely adopted in several critical domains because it facilitates human life and improves quality of service, its security and privacy issues remain a major challenge. As a relief, an anomaly-based Intrusion Detection System (IDS) can be deployed as a security function to safeguard IoT networks from a diverse range of cyber-attacks. In this paper, an anomaly-based IDS is proposed to overcome a diverse range of cyber-attacks in IoT environments. The proposed method uses the kernel principal component analysis technique to minimize the dimension of data features and to improve the anomaly detection performance. We employ the kernel extreme learning machine to determine whether the traffic flow is benign or malicious for binary classification, and to classify the group of attacks to its specific type for multiclass classification. To validate the efficacy of the proposed anomaly detection method, two modern datasets are used to evaluate and analyze the performance results. The evaluation results demonstrate that the proposed anomaly detection approach can effectively improve the detection efficiency and significantly enhance the detection performance results in terms of accuracy rate, specificity rate, sensitivity rate, F1-score, and the area under curve.

The conception of a Computer-Aided Diagnosis system (CAD) using Artificial Intelligence (AI) is a hot topic in the domain of medical diagnosis. Recently, many approaches have been developed. In the proposed work, a novel classification technique from mammograms based on Kernel Extreme Learning Machine (KELM) and Kernel Partial Least Square (KPLS) method is introduced. The suggested algorithm first used the KPLS algorithm to extract features from the images. The extracted characteristics were then sent to the KELM classifier. In order to improve the generalization of the proposed approach, the cross-validation strategy was used. The simulation results were tested on the Mammographic Image Analysis Society (MIAS) dataset and measured using accuracy, F score, sensitivity, and specificity analysis. These results were compared to existing approaches tested on the same dataset and it was observed that the proposed work is the most efficient.

An automated computer‐assisted medical diagnosis that combines latest medical approaches and the advanced machine learning algorithms is a very crucial multidisciplinary technology, generating correct and noninvasive diagnoses of multiple diseases like breast cancer. The work proposed in this article focuses on the development of a biomedical computer‐assisted diagnosis model that can classify digital mammography as normal (healthy) or abnormal, and further, as malignant or benign. The proposed approach employs the discrete Chebyshev transform to extract the features. Then, the kernel principal component analysis technique is used to extract the discriminating features from the original feature vector. Subsequently, an optimized kernel extreme learning machine is proposed as a classifier to detect the tumors present in the mammographic images. Because the efficiency of the proposed classifier depends on its characteristic kernel variable, the main idea of the present work is to choose the most appropriate features from the downsized feature set and simultaneously obtain the optimized value of the aforementioned parameter. To validate the efficiency of the proposed work, the proposed scheme is performed on two publicly available data sets, namely the Mammographic Image Analysis Society data set and the INbreast data set. From the experimental analysis and its results, it is showed that for both normal–abnormal and malignant–benign classification, the proposed approach results in accuracy of 100% for the first data set. However, in the case of malignant–benign classification, the proposed approach gives an accuracy of 99.93% for the second data set. Further, it is also observed that the proposed approach exhibits highest performance as compared to that of the other approaches. Additionally, the ANOVA test is evaluated to demonstrate that the achievement of the proposed approach is significantly good than that of the other existing approaches.

Cancer is one of the deadliest diseases for human health. The classification of cancers poses many challenges in biomedical research because it allows an accurate and effective diagnosis and guarantees the success of medical treatments. In this paper, a new reduced Computer-Aided Diagnosis (CAD) system is implemented under the MATLAB (version R2016a) environment to classifying four cancer subtypes. The results of the experiment are carried out with four sets of baseline data on the expression of cancer genes. To validate the proposed CAD system, different performance metrics such as sensitivity, specificity, accuracy, and F-Score are measured. The experimental analysis justifies the effectiveness of the proposed model and, therefore, this model can be considered as an effective tool to help radiologists for a better diagnosis.

Certain applications of wireless sensor networks require that the sensor nodes should be aware of their position relative to the sensor environment. Generally in the applications of positioning in the internet of things (IoT), there is a deficiency of localization precision and concrete results. It is really important to maintain high-efficient localization schemes for the Internet of things, especially for wireless sensor networks. For that, an improved DV-hop algorithm is proposed in this paper to move for more accurate results based on the least square method. Therefore, the weight coefficient is calculated between an anchor node and the other anchor nodes using the mean square method. Then, this weighting coefficient, the hop size, will be applied between the unknown nodes and the anchor nodes in order to determine the distances. The computed hop-size average significantly enhances the positioning accuracy which is approved by the experiments that explain how suitable this improved Last-Square DV-hop Algorithm is for localization in WSN.

Localization of cerebral electrical activity of emotional states on the basis of electrophysiological recordings is an important area of investigation in recent years. This field was explored to locate the sources of the emotions in the cortex. The theory that every emotion can have a unique trigger center in the cortex was followed. A precise and accurate method was used, the Trilateration, recognized in the GPS networks, which deduces the points of interest from the distances. This method gave the exact coordinates of the generating points of emotions in the surface area of cortex under the influence of modulating thalamic nuclei. It was found that the energies are stronger in the occipital and parietal part of the brain. Moreover the frontal part plays the role of inhibitor and stimulator of emotions.

In this paper, we address the problem of stabilization of switched linear systems. The idea is to look for a state feedback control law using evolutionary algorithms (EA) in order to assure the stability of the switched linear systems under control constraints. In some cases when states are not available and only outputs are measurable, the previous method is applied to design an output feedback controller which stabilizes the system. Both stabilizing controllers are developed using deferential evolution and genetic algorithm. Two numerical examples illustrate our proposed theory and point out the effectiveness of our proposed approaches.