The ease of use and flexibility provided by drones or Unmanned Aerial vehicles (UAV) is attracting different industries and researchers across domains (e.g., delivery, agriculture, security, etc.). Although maintaining a reliable and secure command and control communication channel is still an open challenge and primary limitation for using drones. Satellite and 5G are considered viable solutions for drone communication. In this survey paper, we have explored specifications and proposed enhancements in cellular technology specified by 3GPP to command and control UAVs. It also describes the required network Quality of Service (QoS) parameters for drone communication. Such as end-to-end latency to send and receive a command and control message (C2), reliability, and message size. Along with these, it also emphasizes defining the reliability in terms of communication and navigation of UAVs, based on cellular technology 5G additional investigation and standardization should be executed.
The increase in computing power and integration of specialized hardware for Artificial Intelligence (AI) acceleration like Tensor Processing Units (TPU) enable complex machine learning at edge devices in the Internet of Things (IoT). However, wireless portable systems are limited in computing power and battery lifetime. To increase the battery lifetime of edge devices and accelerate inference of IoT systems, many developments focus on combining or outsourcing AI algorithms to a cloud via wireless links e.g. wireless LAN IEEE 802.11ac or mobile network 4G/5G. Due to limitations of restricted wireless transmissions in rural areas mainly below 50 MBit/s, resulting longer transfer times can significantly affect inference latency and energy consumption from the perspective of the IoT edge device and deteriorate the response time of the application. In this work, we provide a prototype setup for image processing via Convolutional Neural Networks (CNN) and investigate inference latency and energy consumption of an IoT edge device with a varying wireless link. The complexity of selected pre-trained CNN models is between 300 MFLOPs to 19.6 GFLOPs where FLOPs are Floating Point Operations. The first experiments address the latency and energy consumption by processing CNN models on the IoT device with and without TPU as edge AI accelerator. Following experiments address the latency and energy consumption on the IoT device in cloud processing mode with and without Graphics Processing Unit (GPU) as cloud AI accelerator. The edge device sends input data and receives the results via wireless link from 1 MBit/s to 50 MBit/s. For CNN models with ≤564 MFLOPs edge processing with AI acceleration performs better than cloud processing regarding latency and energy efficiency. Even for complex CNN models with 7.6 GFLOPs edge processing can be useful at limited wireless link data rates up to 14 MBit/s. Edge processing without AI acceleration is only an option for low complexity ( ≤300 MFLOPs) and low expected wireless link data rates.
Bluetooth Low Energy (BLE) is one of the technology for indoor positioning. The coordinates in indoor positioning are calculated from the radio signal strength, although distance errors occur between the measured and calculated positions. This paper proposes the method for selecting BLE beacons in multi-point positioning for distance error reduction. The proposed method consists of Learning and Calculation Phase. The correct values used in the proposed method are the values measured with the tape measure. The Learning Phase obtains features from the RSSI distribution of each beacon and uses them to compute the coordinates in the Calculation Phase. The Calculation Phase is used for navigation with an allowable distance error of 0.5 [m] based on a person’s shoulder width. The proposed method calculates the typical value from the Learning Phase based on the probability that it fits within the distance error range of 0.5 [m]. The calculated position uses the logarithmic approximation to convert the beacon’s radio signal strength to the distance. The experiment for measuring the distance error compares the proposed method calculated, the existing method calculated by two-dimensional three-point positioning, and the tape measure-based method. The number of combinations in three-point positioning was also changed to compare the accuracy. The mean distance error compared to the value measured with a tape measure was 1.337 [m] for the existing method and 0.449 [m] for the proposed method. The proposed method is 66.3% less distance error than the existing methods.
This paper addresses the development of a persuasive IoT system for stress detection and management in students during classroom situations. An emotion-aware persuasive architecture is developed with four modules: Context Acquisition, Context Manager, Persuasion Manager and Context-Aware Applications. By using the galvanic skin response biomarker, the real-time stress level is measured by the wearable wristband Empatica E4. The data, processed and classified on discrete stress levels from 0 to 5, is sent to the context module that identifies situations of interest where the students need positive reinforcement from the persuasive system. Based on the situation of interest and the user’s profile, the persuasive module composes personalized persuasive messages displayed in a mobile application. The persuasive system was evaluated through an exploratory study during a class session, with encouraging results in detecting stress levels and the positive effect of persuasive messages on students.
Atrial fibrillation (AF) is a cardiac arrhythmia occurring when the atria lose their normal rhythm causing the heart to beat erratically. The estimated number of individuals with atrial fibrillation globally in 2010 was 33.5 million. Despite continued research in this area there is no universal standard for detecting atrial fibrillation. The majority of published detectors rely on manual classification techniques that are implemented on standalone devices. This paper proposes a dual convolutional neural network (CNN) based AF detection system. The proposed system transforms 5 s windows of electrocardiogram data to two-dimensional images via a stationary wavelet transform to serve as CNN inputs. The dual CNN system implements a model tailored for an IoT gateway device to prescreen arrhythmia cases locally. Less obvious arrhythmia cases are transferred to a secondary model hosted on a cloud server for further prediction. Local classification of AF reduces the overheads for cloud storage capacity and transfer of data. The proposed runtime system ultimately received an F1 score of 0.94 when evaluated using previously unseen data.
The Internet of Things (IoT) is the enabler for new innovations in several domains. It allows the connection of digital services with real, physical entities. These entities are devices of different categories and range in size from large machinery to tiny sensors. In the latter case, devices are typically characterized by limited resources in terms of computational power, available memory and sometimes limited power supply. As a consequence, the use of security algorithms requires expert knowledge in order for them to work within the limited resources. That means to find a suitable configuration for the algorithms to perform properly on the device. On the other side, there is the desire to protect valuable assets as strong as possible. Usually, security goals are captured in security policies, but they do not consider resource availability on the involved device and their consumption while executing security algorithms. This paper presents a resource aware information exchange model and a generation tool that uses high-level security policies as input. The model forms the conceptual basis for an automated security configuration recommendation system.
Seeking a potential of low carbon-based energy use for additive manufacturing, we present a preliminary experimental test using open source IoT tools on FDM (Fused Deposit Modelling) type of 3D printing. In our test we determine and categorize the electricity consumption of processes of a commercial grade FDM printer using a custom-built energy monitor. Our tests indicate that this model of FDM type 3D printer consumes between 22
The design of decentralized learning algorithms is important in the fast-growing world in which data are distributed over participants with limited local computation resources and communication. In this direction, we propose an online algorithm minimizing non-convex loss functions aggregated from individual data/models distributed over a network. We provide the theoretical performance guarantee of our algorithm and demonstrate its utility on a real life smart building.
One of the challenges in the Computing Continuum paradigm is the optimal distribution of the generated tasks between the devices in each layer (cloud-fog-edge). In this paper, we propose to use Reinforcement Learning (RL) to solve the Task Assignment Problem (TAP) at the edge layer and then we propose a novel multi-layer extension of RL (ML-RL) techniques that allows edge agents to query an upper-level agent with more knowledge to improve the performance in complex and uncertain situations. We first formulate the task assignment process considering the trade-off between energy consumption and execution time. We then present a greedy solution as a baseline and implement our two RL proposals in the PureEdgeSim simulator. Finally, several simulations of each algorithm are evaluated with different numbers of devices to verify scalability. The simulation results show that reinforcement learning solutions outperformed the heuristic-based solutions and our multi-layer approach can significantly improve performance in high device density scenarios.
This paper provides an overview of the Advanced Threat Intelligence Orchestrator in assisting organizations and society’s first responders in managing, prioritizing, and sharing information related to cyber security incidents. In order to accomplish this, the capabilities and benefits of security, orchestration, automation, and response (SOAR) systems, on which Orchestrator is based, were promoted. The results of this survey conducted as part of the IRIS EU-funded project to protect Internet of Things (IoT) and Artificial Intelligence (AI)-driven ICT-enabled systems from cyber threats and attacks on their privacy facilitating SOC/CSIRTs/CERTs. In this context, the tool is explored in methods of orchestrating and automating cyber security processes and routines. The open-source tool that was chosen for the creation of Advanced Threat Intelligence Orchestrator was SHUFFLE. SHUFFLE gives a wide variety of functionalities as it can be integrated with numerous tools and APIS. Furthermore, the provision of schematic workflows with action steps makes the stakeholders’ interface more intuitive.
The wide deployment of Machine Learning models is an essential evolution of Artificial Intelligence, predominantly by porting deep neural networks in constrained hardware platforms such as 32 bits microcontrollers. For many IoT applications, the deployment of such complex models is hindered by two major issues that are usually handled separately. For supervised tasks, training a model requires a large quantity of labelled data which is expensive to collect or even intractable in many real-world applications. Furthermore, the inference process implies memory, computing and energy capacities that are not suitable for typical IoT platforms. We jointly tackle these issues by investigating the efficiency of model pruning techniques under the scope of the single domain generalization problem. Our experiments show that a pruned neural network retains the benefit of the training with single domain generalization algorithms despite a larger impact of pruning on its performance. We emphasize the importance of the pruning method, more particularly between structured and unstructured pruning as well as the benefit of data-agnostic heuristics that preserve their properties in the single domain generalization setting.
Industries across the world produce thousands of confidential images and other data that needs to be secured on a daily basis. With the onset of the information era and the exponential growth of technologies like big data, security of private and confidential information in any form has become a challenge. Many times, these images need to be transmitted over public networks to facilitate wireless monitoring and sharing testing results with remote operators. This transmission needs to be secure to ensure the safety and confidentiality of the images, thus cryptography techniques like image steganography are required. In this prototype, we have designed, developed and tested a Raspberry Pi LAMP web server which stores images locally which are obtained using an ESP32-Cam, and which facilitates transmission using the steganography technique. There is also an option to store these images on any other widely used cloud server, and the transmission can still use steganography encryption for an extra layer of security. The stored images can only be accessed by authorized users.
Crowd counting is of great importance to many applications in various scenarios. Wi-Fi Channel State Information (CSI)-based crowd counting is a highly accurate privacy-conscious method. However, the problem with CSI-based crowd counting is the size and cost of the CSI collecting tool. Most studies benefiting from CSI collection use laptops with specific Network Interface Cards (NICs). The size and cost of the laptops restrict the practicability of such systems and limit active repositioning and mobility of the devices. This research aims to realize highly accurate CSI-based crowd counting using only one pair of lightweight and low-cost IoT devices. The devices are very agile and can easily be deployed even in space-limited environments. However, they have the disadvantage of poor data transportation compared to laptops. We compensate for this drawback by adjusting the deployment location, using multiple preprocessing methods depending on the situation, and standardizing the data for each subcarrier. We conducted evaluations of crowd counting in two representative scenarios. For the scenario of crowd sizes of 0, 1, 2, and 3 persons, when we used a weighted moving average (WMA) filter and phase sanitization as the preprocessing methods, the accuracy was 70.3%. When we used percentage of nonzero elements (PEM) and a moving average (MA) filter as the preprocessing methods, the accuracy was 84.6%. For the scenario of crowd sizes of 0, 5, 10, 15, and 20 persons, when we used a WMA filter and phase sanitization as the preprocessing methods, the accuracy was 76.5%. When we used PEM and a MA filter as the preprocessing methods, the accuracy was 75.9%. We found that the appropriate preprocessing method differs between the case of a small number of people and the case of a large number of people.
The recent focus on deep learning accuracy ignored economic and environmental cost. Introduction of Green AI is hampered by lack of metrics that balance rewards for accuracy and cost and thus improve selection of best deep learning algorithms and platforms. Recognition and training efficiency universally compare deep learning based on energy consumption measurements for inference and deep learning, on recognition gradients, and on number of classes. Sustainability is assessed with deep learning lifecycle efficiency and life cycle recognition efficiency metrics that include the number of times models are used.
In this paper, we propose a novel approach in smart farming with the deployment of centrally controlled IoT-scaring devices in meadows with the goal to reduce the killing of roe deer fawn during haymaking. These deaths are due to fawns not actively avoiding threats in their first two weeks of life, employing a defensive strategy of hiding scentless and motionless in order to avoid predation instead. Currently, they are searched and removed from areas to be mowed by hand. Our approach allows for a reduction of the labour required in advance of a scheduled mowing. During field tests, the effectiveness of the devices has been shown in northern Germany.
This paper summarizes the key findings of a qualitative study based on feedbacks of experts from the wind industry followed by a demo case of blockchain technology. This study includes investigation on mapping of supply chain and commodity products related end-to-end life cycle associated data and operational events. Furthermore, identification and blueprinting of the requirements have been pursued for enabling traceability at various stages of their life cycle utilizing blockchain. In this study context, blockchain offers digital traceability of operational events and associated data sharing with complete immutability, ownership, confidentiality, trust, and transparency across distributed supply chain which comprises of multiple stakeholders. In addition, digital technology intervention like IoT has been leveraged to support quality of operations in quantitative manner through real time data driven digitized operations. Thereby, providing economy of scale over operations execution on commodity products in wind industry. This study has focused only on bolts and fasteners associated commodity products and related supply chain. However, this provides a steppingstone foundation for future which can be scaled and mapped to any other commodity product and related supply chain in wind industry. Finally, this study also presents a demonstrator developed in a controlled lab environment to demystify the use of blockchain technology in related manufacturing and supply chain setups of the wind industry.
The flexible PCB medical device developed at our research lab calculates a single-channel EOG. We develop an infrastructure for our device, including an IoT structure for capturing data. As well as an algorithm that can detect Sleep Stages using EOG data from our device. Previous attempts at classifying sleep always use data from double-channel EOG data. Initially, we used a labelled sleep dataset from the University of Wisconsin to train our neural network. We then apply transfer learning to the sleep classifier with data extracted from our device. Overall, we were able to successfully create a model with data from the medical device and obtain a 81.19% sleep stage classification accuracy.
This paper presents the activities of the Applied IoT Lab at the Department of Computer Science and Media Technology, Linnaeus University (LNU), Kalmar, Sweden. The lab is actively engaged in IoT-based educational programs, including a series of workshops and pilot cases. The lab is funded by the European Union and two Swedish counties – Kalmar and Kronoberg. The workshops and pilot cases are part of the research project named IoT Lab for Small and Medium-sized Enterprises (SMEs). One of the lab’s main objectives is to strengthen and support local companies with IoT. The project IoT Lab for SMEs also aims to spread knowledge and inspire the local community about the possibilities of using IoT technologies by organizing open lab days, in-depth lectures, and seminars. This paper introduces Applied IoT Lab at LNU, its educational programs, and industry-academic cooperation, including workshops and a number of ongoing pilot cases.
The ERATOSTHENES project is driven by recent security challenges of IoT networks being today embedded into our day to day lives. The high increase of connected devices, their inhomogeneous nature, high penetration, as well as different manufacturing and vendor characteristics have created a vast attack surface that is prone to increase in the next years. This has already created challenges such as: confidentiality access control, privacy for users and things, devices’ trustworthiness and compliance that require lifecycle considerations of IoT devices and networks. ERATOSTHENES will devise a novel distributed, automated, auditable, yet privacy-respectful, Trust and Identity Management Framework intended to dynamically and holistically manage the lifecycle of IoT devices, strengthening trust, identities, and resilience in the entire IoT ecosystem, supporting the enforcement of the NIS directive, GDPR and Cybersecurity Act. This publication positions the project into the internet of things and applications and describes the project concept, requirements, first architectural decisions and outcomes.
Ports are essential nodes in global maritime trade. As such, their efficency is key to ensure sustainable supply chains across the world. Current studies point interoperability and data integration as the next milestones to achieve efficient, smart ports. DataPorts aims at covering those gaps, delivering an industrial data platform bearing in mind seaports’ involved actors’ needs. With DataPorts, transportation and logistics companies will leverage the current data deluge to offer cognitive services. This paper describes the technical design of such a platform and how it enables the acquisition, homogenization, and processing of the heterogeneous data, judiciously handled to generate advanced data-exhaustive services. Finally, it presents a practical usage example aimed to improve the business processes in the port of Valencia.
