Multicopter drones are becoming a key platform in several application domains, enabling precise on-the-spot sensing and/or actuation. We focus on the case where the drone must process the sensor data in order to decide, depending on the outcome, whether it needs to perform some additional action, e.g., more accurate sensing or some form of actuation. On the one hand, waiting for the computation to complete may waste time, if it turns out that no further action is needed. On the other hand, if the drone starts moving toward the next point of interest before the computation ends, it may need to return back to the previous point, if some action needs to be taken. In this paper, we propose a learning approach that enables the drone to take informed decisions about whether to wait for the result of the computation (or not), based on past experience gathered from previous missions. Through an extensive evaluation, we show that the proposed approach, when properly configured, outperforms several static policies, up to 25.8%, over a wide variety of different scenarios where the probability of some action being required at a given point of interest remains stable as well as for scenarios where this probability varies in time.
The integration of Internet of Things (IoT) applications in our daily lives has led to a surge in data traffic, posing significant security challenges. IoT applications using cloud and edge computing are at higher risk of cyberattacks because of the expanded attack surface from distributed edge and cloud services, the vulnerability of IoT devices, and challenges in managing security across interconnected systems leading to oversights. This led to the rise of ML-based solutions for intrusion detection systems (IDSs), which have proven effective in enhancing network security and defending against diverse threats. However, ML-based IDS in IoT systems encounters challenges, particularly from noisy, redundant, and irrelevant features in varied IoT datasets, potentially impacting its performance. Therefore, reducing such features becomes crucial to enhance system performance and minimize computational costs. This paper focuses on improving the effectiveness of ML-based IDS at the edge level by introducing a novel method to find a balanced trade-off between cost and accuracy through the creation of informative features in a two-tier edge-user IoT environment. A hybrid Binary Quantum-inspired Artificial Bee Colony and Genetic Programming algorithm is utilized for this purpose. Three IoT intrusion detection datasets, namely NSL-KDD, UNSW-NB15, and BoT-IoT, are used for the evaluation of the proposed approach. Performance analysis is conducted using various evaluation metrics such as accuracy, sensitivity, specificity, and False Positive Rate (FPR) are employed, while the cost of the IDS system is assessed based on computational time. The results are compared with existing methods in the literature, revealing that the IDS performance can be enhanced with fewer features, consequently reducing computational time, through the proposed method. This offers a better performance-cost trade-off for the IDS system.
As more devices connect to the internet, it becomes crucial to address their limitations and basic security needs. While much research focuses on utilizing ML and DL to tackle security challenges, there is often a tendency to overlook the practicality and feasibility of implementing these methods in real-time settings. This oversight stems from the constrained processing power and memory of certain devices (IoT devices), as well as concerns about the generalizability of these approaches. Focusing on the detection of DNS-tunneling attacks in a router as a case study, we present an end-to-end process designed to effectively address these challenges. The process spans from developing a lightweight DNS-tunneling detection model to integrating it into a resource-constrained device for real-time detection. Through our experiments, we demonstrate that utilizing stateless features for training the ML model, along with features chosen to be independent of the network configuration, leads to highly accurate results. The deployment of this carefully crafted model, optimized for embedded devices across diverse environments, resulted in high DNS-tunneling attack detection with minimal latency. With this work, we aim to encourage solutions that strike a balance between theoretical advancements and the practical applicability of ML approaches in the ever-evolving landscape of device security.
Smart farming and precision agriculture represent game-changer technologies for efficient and sustainable agribusiness. Miniaturized palm-sized drones can act as flexible smart sensors inspecting crops, looking for early signs of potential pest outbreaking. However, achieving such an ambitious goal requires hardware-software codesign to develop accurate deep learning (DL) detection models while keeping memory and computational needs under an ultra-tight budget, i.e., a few MB on-chip memory and a few 100s mW power envelope. This work presents a novel vertically integrated solution featuring two ultra-low power System-on-Chips (SoCs), i.e., the dual-core STM32H74 and a multi-core GWT GAP9, running two Stateof-the-Art DL models for detecting the Popillia japonica bug. We fine-tune both models for our image-based detection task, quantize them in 8-bit integers, and deploy them on the two SoCs. On the STM32H74, we deploy a FOMO-MobileNetV2 model, achieving a mean average precision (mAP) of 0.66 and running at 16.1 frame/s within 498mW. While on the GAP9 SoC, we deploy a more complex SSDLite-MobileNetV3, which scores an mAP of 0.79 and peaks at 6.8 frame/s within 33mW. Compared to a top-notch RetinaNet-ResNet101-FPN full-precision baseline, which requires 14.9x more memory and 300x more operations per inference, our best model drops only 15% in mAP, paving the way toward autonomous palm-sized drones capable of lightweight and precise pest detection.
The escalating volumes of textile waste globally necessitate innovative waste management solutions to mitigate the environmental impact and promote sustainability in the fashion industry. This paper addresses the inefficiencies of traditional textile sorting methods by introducing an autonomous textile analysis pipeline. Utilising robotics, spectral imaging, and AI-driven classification, our system enhances the accuracy, efficiency, and scalability of textile sorting processes, contributing to a more sustainable and circular approach to waste management. The integration of a Digital Twin system further allows critical evaluation of technical and economic feasibility, providing valuable insights into the sorting system’s accuracy and reliability. The proposed framework, inspired by Industry 4.0 principles, comprises five interconnected layers facilitating seamless data exchange and coordination within the system. Preliminary results highlight the potential of our holistic approach to mitigate environmental impact and foster a positive shift towards recycling in the textile industry.
Many AI platforms, including traffic monitoring systems, use Federated Learning (FL) for decentralized sensor data processing for learning-based applications while preserving privacy and ensuring secured information transfer. On the other hand, applying supervised learning to large data samples, like high-resolution images requires intensive human labor to label different parts of a data sample. Multiple Instance Learning (MIL) alleviates this challenge by operating over labels assigned to the 'bag' of instances. In this paper, we introduce Federated Multiple-Instance Learning (FedMIL). This framework applies federated learning to boost the training performance in video-based MIL tasks such as vehicle accident detection using distributed CCTV networks. However, data sources in decentralized settings are not typically Independently and Identically Distributed (IID), making client selection imperative to collectively represent the entire dataset with minimal clients. To address this challenge, we propose DPPQ, a framework based on the Determinantal Point Process (DPP) with a quality-based kernel to select clients with the most diverse datasets that achieve better performance compared to both random selection and current DPP-based client selection methods even with less data utilization in the majority of non-IID cases. This offers a significant advantage for deployment on edge devices with limited computational resources, providing a reliable solution for training AI models in massive smart sensor networks.
Graph-based or multidimensional blockchains have been proposed to improve the scalability and efficiency of existing blockchain applications. However, when implemented in mobile Internet of Things (mIoT) networks, these blockchain systems can frequently split and merge and cause the merging algorithm to process a large number of similar blocks (a block is similar or identical when it exists once in multiple blockchains). The presence of similar blocks hinders the merging process, as it consumes time and computational resources to scan, validate, and potentially merge similar blocks. This paper presents an efficient approach for merging graph-based or multidimensional blockchains in mIoT networks by avoiding similar blocks and effectively detecting and merging new blocks that were created after the split. Our proposed merging algorithm employs depth-first search and Merkle tree techniques to minimize the time and computational resources spent on identical blocks. Finally, we evaluate the performance of our method in highly mobile networks and demonstrate that it can execute the merge with a more than 72% reduction in time in comparison to merging algorithms without block similarity handling.
The dynamic nature of many IoT ecosystems poses unique challenges to the efficacy of IoT ML-based applications. One such challenge is data incompleteness. Furthermore, most IoT systems are severely power-constrained. It is important that we build IoT-based ML systems that are robust against data incompleteness while simultaneously being energy efficient. This paper presents an empirical study of SECOE a recent technique for alleviating data incompleteness in IoT with respect to its energy bottlenecks. Towards addressing the energy bottlenecks of SECOE, we propose ENAMLE a proactive, energy-aware technique for mitigating the impact of concurrent missing data. ENAMLE is unique in the sense that it builds an energy-aware ensemble of sub-models, each trained with a subset of sensors chosen carefully based on their correlations. Furthermore, at inference time, ENAMLE adaptively alters the number of the ensemble of models based on the amount of missing data rate and the energy-accuracy trade-off. ENAMLE's design includes several novel mechanisms for minimizing energy consumption while maintaining accuracy. Through experimental studies, we demonstrate the energy efficiency of ENAMLE and its ability to alleviate sensor failures.
Sharing and joint processing of camera feeds and sensor measurements, known as Cooperative Perception (CP), has emerged as a new technique to achieve higher perception qualities. CP can enhance the safety of Autonomous Vehicles (AVs) where their individual visual perception quality is compromised by adverse weather conditions (haze as foggy weather), low illumination, winding roads, and crowded traffic. To cover the limitations of former methods, in this paper, we propose a novel approach to realize an optimized CP under constrained communications. At the core of our approach is recruiting the best helper from the available list of front vehicles to augment the visual range and enhance the Object Detection (OD) accuracy of the ego vehicle. In this two-step process, we first select the helper vehicles that contribute the most to CP based on their visual range and lowest motion blur. Next, we implement a radio block optimization among the candidate vehicles to further improve communication efficiency. We specifically focus on pedestrian detection as an exemplary scenario. To validate our approach, we used the CARLA simulator to create a dataset of annotated videos for different driving scenarios where pedestrian detection is challenging for an AV with compromised vision. Our results demonstrate the efficacy of our two-step optimization process in improving the overall performance of cooperative perception in challenging scenarios, substantially improving driving safety under adverse conditions. Finally, we note that the networking assumptions are adopted from LTE Release 14 Mode 4 side-link communication, commonly used for Vehicle-to-Vehicle (V2V) communication. Nonetheless, our method is flexible and applicable to arbitrary V2V communications.
Data collection is a strenuous and time-consuming process, particularly when aiming to build large datasets that are essential for training deep models effectively. While curating such datasets in controlled environments requires meticulous design and annotation, adding complexity to the process, the prevalence of mobile devices has led to the collection of extensive crowdsourced datasets reflecting diverse user behaviours. However, real-world datasets often suffer from incomplete, noisy, or incorrect annotations that introduce significant variability in user behaviour within each class, hindering model learning and generalisation. Despite various methods in these research areas attempting to mitigate this problem, there is a clear lack of explainable solutions to discerning real-world model deployment while providing underlying information about incoming data. To address this challenge, we propose SeMEDA, a Semantic Mismatch Estimation and Dataset Alignment approach which automates the alignment of datasets by establishing a cross-domain mapping that represents incoming data from a target domain through the scope of a model trained on a controlled dataset. SeMEDA identifies and addresses four key levels of semantic mismatch, enhancing the curation of cleaner datasets with trustworthy labels without the need for laborious data analysis and expert annotation. We showcase our proposed approach through two datasets: the watchHAR dataset, which was collected in controlled laboratory conditions, and the ExtraSensory dataset, which was collected in-the-wild, boosting performance accuracy in target space from 45% to over 90%.
The IoT device marketplace has seen a large boom over the past decade. Myriad tech companies introduce IoT products to the market to capitalize on this demand. However, most IoT products have hidden security vulnerabilities, and these back-doors could be exploited by adversaries, which could lead to privacy-sensitive data leakage, unauthorized access or even physical harm to IoT device users. The event spoofing attack is one such attack where the adversary compromises one IoT device in the smart home ecosystem and sends spoofed event updates to cause undesired and unexpected outcomes of other IoT devices. Fortunately, previous research in this area has shown that it is possible to identify these attacks by analyzing the sensor readings of IoT devices to validate the status updates of other IoT devices. This paper proposes a novel Recurrent Neural Network based auto-encoder architecture that performs better than the current state-of-the-art in terms of detection accuracy, false positive rate and training time. In addition, the auto-encoder output is easily interpretable; therefore, subsequent identification of misbehaving IoT devices is more accessible for cybersecurity analysts.
It is widely acknowledged that real implementation stands out as the most accurate performance evaluation technique, surpassing simulation-based and analytical-based approaches. However, the inherent challenges and costs associated with real implementation pose significant barriers to its widespread adoption. While several testbeds have successfully navigated these challenges for ad-hoc communication, existing solutions often come with drawbacks, such as high costs, dedication to specific protocols or hardware, and limitations in terms of mobility and AI compatibility. This paper introduces a novel, versatile testbed designed for the performance evaluation of mobile ad-hoc networks (MANETs). Noteworthy for its cost-effectiveness, modularity, mobility, and AI compatibility, this testbed supports non-homogeneous nodes and operates seamlessly on any system with a Linux operating system (OS). It is complemented by a user-level Application Programming Interface (API) that facilitates the implementation of various protocols on the testbed by separating the implementation from the OS functionality 1 . To demonstrate its practicality, we implemented the Ad hoc On-Demand Distance Vector (AODV) routing protocol, a complex and well-known example, and compared its results with those obtained from the network simulator ns-3.
IoT smart applications must deal with the inherently distributed nature of IoT infrastructures that may vary among different deployments. For example, in urban environments, some deployments may require distributed edge nodes to support the sensors and actuators, while others may skip this processing stage and go directly to the cloud. We propose the concept of an IoT Computing Continuum, or IoTinuum, with multiple computing stages spread over the distributed IoT infrastructure. We present the modeling and implementation of two urban use cases for a 6-stage IoTinuum: smart drone delivery and smart structural monitoring. In our experience, the IoTinuum improves the understanding of function placement in application development and reveals important performance tradeoffs.
This paper provides a novel solution for developing a virtual keyboard and mouse (VKM) system that is easily manageable and portable. The traditional keyboards and mouse devices take up valuable desk space and are not easily customizable to different languages. On-screen keyboards and 3-D cameras are alternatives, but they also have drawbacks. Our proposed method makes use of computer vision techniques and calls for a mini-projector and a web camera as necessary hardware. The system tracks hand keypoints to detect real-time touch events and uses the Mediapipe tool to detect hands and keystrokes. The mouse functionality is also implemented by monitoring the finger hovering. Through experimentation, we show that our VKM solution can provide an accuracy of >90% for detecting the correct keystroke, with a typing speed of similar to 55 letters/min.
The Internet of Things (IoT) has revolutionized numerous domains, but security and privacy remain significant concerns. Massive amounts of IoT data poses challenges for a centralized IDS. Distributed solutions, particularly using Machine Learning (ML) and Deep Learning (DL) methods, demand substantial data and processing power, thus impractical for resource-constrained IoT devices.This study employs three ML, one DL, and five One-Class Classification (OCC) methods for anomaly detection across five IoT datasets. While supervised ML and DL methods demonstrate superior performance, only some OCC methods exhibit comparable performance. All ML and DL methods show consistent performance on all different datasets, but this seamless consistency is not observed among OCCs. DT stands out as the most accurate algorithm among all, and DROCC and LOF stand out as the most accurate and consistent OCCs. In terms of execution time, DT is the fastest among all methods, and DROCC and LOF are the second and third fastest among OCCs. Conversely, among the most time-consuming algorithms, OCSVM is the most time-intensive, followed by DeepSVDD and DNN.
Intelligent transportation systems (ITS) use the latest technologies for real-time traffic control and monitoring to ensure efficient traffic management and reduce the risks of traffic accidents. The microscopic level of traffic modeling is the most appropriate level for controlling and monitoring the interaction between vehicles based on car-following scenarios. However, the data retrieved from sensor networks can be affected by measurement errors, and consequently the implementation of appropriate mechanisms to overcome their propagation to the control system is mandatory. This paper aims to analyse the current research in the calibration of car-following models and provide valuable insights of recent developments in this field. To achieve this goal, VOSviewer has been chosen as a visualisation tool to create bibliographic maps based on the output from the well-known scientific database Clarivate Analytics Web of Science (WoS). The maps obtained provide a visual representation of the main institutions involved in this field of research and identify the research interests based on author and indexing keywords. Furthermore, this paper analyses the top five clusters identified based on the analysis of co-occurrence keywords, presenting discussions about the connections existing within these clusters.
This work addresses the challenge of text-based person re-identification (re-ID) on resource-constrained embedded devices, a critical component in modern surveillance systems. Text-based person re-ID involves using textual descriptions to search persons across multiple camera views. Implementing such algorithms on embedded devices such as smart cameras is challenging due to limited memory and computational constraints. In this work, we propose TextReIDNet, a lightweight person re-ID model designed explicitly for embedded devices. Compared to state-of-the-art models, TextReIDNet aims at an optimal balance between person re-ID accuracy and computational efficiency, thus making it well-suited for low-resource devices. With the smallest model size of only 32.29 million parameters, TextReIDNet achieves a competitive 52.76% and 35.71% top-1 accuracy on the CUHK-PEDES and RSTPReid datasets, respectively. We implemented TextReIDNet on the Jetson Nano board to demonstrate its capability for embedded deployments. On average, TextReIDNet requires 1.13ms to process a text and 30.92ms for an image.
Intelligent Transportation Systems (ITS) play a pivotal role in shaping the foundation of smart cities, providing data-driven solutions for traffic management, prediction, and safety. However, these applications often face a significant challenge - data scarcity. Insufficient data limits the effectiveness of machine learning models in the context of ITS. To address this issue, this paper presents a novel data augmentation solution using Generative Adversarial Networks (GANs). By collecting sensor-based traffic speed data with contextual labels and training a GAN-based model to generate realistic traffic data for specific days and times, this research successfully proposes a solution to the problem of data scarcity. The generated data undergoes comprehensive qualitative and quantitative evaluations, demonstrating its potential to enhance ITS applications. Furthermore, the generated data is utilized to augment the training data for multiple traffic prediction models, effectively enhancing their performance. This approach opens new avenues for the development of intelligent and sustainable transportation systems, ultimately contributing to the advancement of smarter and more resilient cities.
Fog computing, an evolution of cloud computing, has become increasingly popular for its ability to lessen the burden of such a centralized computing paradigm by distributing tasks generated by IoT across fog layers. Effectively managing real-time, delay-sensitive, and diverse IoT applications to enhance the Quality-of-Experience (QoE) presents significant challenges due to the dispersed nature and limited resources of fog nodes. Previous studies in fog computing task offloading have typically focused on either energy consumption or service delay. This paper introduces an optimization framework for task offloading within fog computing environments that aims to balance improved user QoE with reduced energy consumption, employing Mixed-Integer Linear Programming (MILP). Given the NP-hard nature of this framework, we have devised a Deep Q-Learning (DQL) based model for task offloading, termed ELTO-DQL, which aims for near-optimal solutions in polynomial time. Experimental results indicate that the ELTO-DQL model enhances energy efficiency and QoE by up to 19% and 15% respectively, outperforming contemporary benchmarks.
The advent of 5G networks has brought about a significant transformation in contemporary society, highlighting notable growth, especially in the real-time video transmission sector but also in streaming services. However, users still face issues with inadequate video quality, particularly when the transmission characteristics of 5G networks vary. Within this context, this paper presents an Artificial Intelligence (AI) model for predicting the video quality delivered to the end user. The proposed model considers both static and dynamic scenarios in vehicular devices, recognizing the importance of understanding content delivery according to the user’s situation. This work’s proposal uses real network data from 5G network measurements, enabling the optimization of user experience in dynamic 5G environments. The results demonstrate that the proposal can contribute to the continuous improvement of video quality delivery in the context of mobile networks.
