Low-end Devices Research Articles

SVM-Based Task Admission Control and Computation Offloading Using Lyapunov Optimization in Heterogeneous MEC Network

Integrating device-to-device (D2D) cooperation with mobile edge computing (MEC) for computation offloading has proven to be an effective method for extending the system capabilities of low-end devices to run complex applications. This can be realized through efficient computing data offloading and yet enhanced while simultaneously using multiple wireless interfaces for D2D, MEC and cloud offloading. In this work, we propose user-centric real-time computation task offloading and resource allocation strategies aiming at minimizing energy consumption and monetary cost while maximizing the number of completed tasks. We develop dynamic partial offloading solutions using the Lyapunov drift-plus-penalty optimization approach. Moreover, we propose a task admission solution based on support vector machines (SVM) to assess the potential of a task to be completed within its deadline, and accordingly, decide whether to drop from or add it to the user’s queue for processing. Results demonstrate high performance gains of the proposed solution that employs SVM-based task admission and Lyapunov-based computation offloading strategies. Significant increase in number of completed tasks, energy savings, and cost reductions are resulted as compared to alternative baseline approaches.

Faster object detector for drone-captured images

Object detection based on computer vision is becoming popular in drone-captured images. However, real-time object detection in unmanned aerial vehicle (UAV) scenarios is a huge challenge for low-end devices. To deal with the problem, we have improved YOLOv3-tiny in the following aspects. First, the label rewriting problem, which is caused by network structure and dataset of YOLOv3-tiny in drone-captured images detection, is very serious. The method of increasing the size of the predicted feature map is used to reduce the ratio of label rewriting. Second, the features of small targets will be reduced in a small feature map, but the context information with large receptive fields in it can improve the performance of small target detection. So we use dilated convolution to expand the receptive field without reducing the size of the feature map. Third, multiscale feature fusion is very helpful for small target detection. The multidilated module is adopted to merge features in earlier layer and deeper layers. Finally, a pretraining strategy combining copy-paste data augmentation method is proposed to learn more features from categories with a small number of samples. We evaluated our model on the VisDrone2019-Det test set. It achieves compelling results compared to the counterparts of YOLOv3-tiny, including ∼86.1 % decline in model size, increasing ∼19.2 % AP50. Although our model is slower than YOLOv3-tiny, it is 2.96 times faster than YOLOv3. The results of experiments verify that our network is more effective than YOLOv3-tiny. It is more suitable for UAV object detection applications on low-end devices.

An Extended Software-Defined Approach for Reprogramming Low-end IoT Devices

The usage of the Software-Defined Sensor Networks (SD-WSNs) as support for applications of the Internet of Things (IoT) has been the focus of recent research to reduce the complexity of configuration and management. This simplification allows greater flexibility for the configuration of these networks in operating time. However, although more flexible than traditional approaches, the existing SD-WSN architectures found in the literature have limitations mainly with regard to the reprogramming of devices in operating time. In this article, we propose a novel approach called Extended Software-Defined Wireless Sensor Network (ESD-WSN), in which the concept of the SDNs is extended to the application layer allowing the reprogramming of low-end devices in operation time without the need for firmware update. The proposed architecture was validated by simulation and an experimental study using very constrained low-end devices. The simulation/experimental results showed the effectiveness, flexibility, and efficiency of the proposal.

SREFBN: Enhanced feature block network for single‐image super‐resolution

Deep learning has assisted the field of single-image super-resolution (SR) in achieving new heights. However, the task of restoring a high-resolution (HR) image from a highly degraded low-resolution (LR) image is sophisticated due to poor image restoration quality. A novel and effective lightweight SR method is presented as super-resolution via an enhanced feature block network (SREFBN) that successfully reconstructs an HR image using a corresponding LR image with a purposed deep residual block. In addition, a novel shared parameters approach in the top-down pathway among low-level feature maps is introduced. The experimental results prove that SREFBN achieves remarkable performance. The presented framework requires lower computational cost and outperforms many state-of-the-art methods. It is also highly adaptable with low-end devices, requiring lower multiplication and adding operations. A trade-off comparison between the number of parameters, execution time, and accuracies is given while also showing different variations of our approach to prove the effectiveness and reliability of the shared parameters. Most importantly, the results indicate that our framework has gained state-of-the-art performance on larger scales 3 and 4. Code is available at https://github.com/curzii23/SREFBN.

Combining System Integrity Verification with Identity and Access Management

Digital transformation and the utilization of Industrial IoT (IIoT) introduces numerous interconnected devices to factories increasing among others the challenge of managing their software versions and giving attackers new possibilities to exploit various software vulnerabilities.
 Factory networks were earlier isolated from the Internet. However, this separation is no longer valid and there can be connections that allow intruders to penetrate into information systems of factories. Another issue is that although factories typically are physically isolated, it is not necessarily safe to assume that physical security is in good shape as the novel supply networks comprise subcontracted activities and temporary work force. Another threat can also arise from unauthorized monitoring of devices and the unauthorized replacement of existing ones.
 Based on the previous, it is crucial that IIoT security should be built into factories of the future (FoF) right from the design phase and even low-end devices need to be supported. Trusted computing concept called remote attestation should be used. Remote attestation allows remote parties to verify the integrity of each system component. System components should include trusted hardware components that can be used to measure executable software. The term measurement means calculating the cryptographic hash of the binary component before passing control to it. Trusted hardware components should also have a mechanism to protect the integrity of the measurement list and cryptographic keys that can be used to sign integrity assertions. The verifier part should have a storage of reference integrity metrics identifying the expected values of these measurements.
 Deploying trusted computing and remote attestation concepts to industrial automation is not straightforward. Even if it is possible to use remote attestation with suitable hardware components, it is not clear how remote attestation should be integrated with various operational technology (OT) industrial automation protocols. Approaches to use remote attestation with existing industrial automation protocols (e.g., OPC UA) is discussed. Advanced identity and access management (e.g., OAuth2, OpenID Connect) can be used to combine integrity measurements with device identity information so that the remote attestation process is triggered by authentication during the first transaction. The focus is on machine-to-machine (M2M) communications with immutable device identities and integrity evidence transfer.

Max-min threshold-based cancelable biometric templates for low-end devices

Biometrics are widely used in security systems, but it has drawbacks such as its nonreusability, being susceptible to getting stolen, and being prone to unauthorized access. These issues can be tackled using cancelable biometrics. In this work, features extracted using Log-Gabor filters are processed according to the proposed max-min thresholding. The resulting binary features are transformed by random projection on a key matrix to create a cancelable template. The method was carefully evaluated using publicly accessible datasets of face near-infrared, palm print, palm vein, dorsal vein, wrist vein, and knuckle print. A significant improvement in performance over many recent methods was observed. Also, the effectiveness of the generated cancelable templates was observed in the general, stolen token, and changeable scenarios. The generated template is 25% of the original input size, which makes it efficient for low-end devices.

Performance Optimization of 3-DOF Application Scene Based on 360-Degree Panoramic Technology with Depth Information

High-quality 3D scenes often show poor rendering effect and insufficient operational performance in low-end devices. Therefore, how to make better use of 360-degree panoramic technology and replace the traditional pure model scene by visual deception is of great significance to the picture quality improvement and performance optimization of low-end equipment. In this paper, a three-degree-of-freedom (3-DOF) application scenario for indoor simulation is used as an example. Under the premise of 360-degree panoramic technology, depth information is obtained through laser ranging, converted into a two-dimensional form of depth image with the help of spherical coordinates, and combined with a transparent information image, the depth image is compressed using DEFLATE compression algorithm, and finally be stored. By optimizing the data through bilinear sampling and with the help of geometric knowledge, the occluded parts are removed or translucent according to the actual situation when the graphics are rendered, and the occlusion relationship is correctly handled in order to achieve a real integration of the virtual environment with the 3D objects. Through multiple data, it is shown that the method possesses significant performance improvement while ensuring geometric realism.

ENABLING TECHNOLOGIES CHALLENGES OF GREEN INTERNET OF THINGS (IOT) TOWARDS SUSTAINABLE DEVELOPMENT IN THE ERA OF INDUSTRY 4.0

The extensive adoption of the Internet of Things (IoT) has increased the carbon footprint on a large scale across the globe. To handle this challenge, scholars and policymakers are making efforts to propose novel energy-efficient solutions to provide a desirable environment for green-IoT (G-IoT). Additionally, further research is required to analyze the G-IoT-related challenges to elucidate the difficulties of its implementation for researchers. Moreover, the GIoT requirements have been considered in different network levels, namely software, hardware, architecture, communication. To present a comprehensive framework to identify the challenges of G-IoT, a survey using literature review and expert’s opinion is carried. Total 23 challenges are taken to evaluate and implement G-IoT technologies towards sustainable development achievements (SDA). Consequently, this article aims to rank and evaluate the challenges to implement the G-IoT towards the SDA. An integrated approach is proposed with stepwise weight assessment ratio analysis (SWARA) and additive ratio assessment (ARAS) under Pythagorean fuzzy sets. As a result, an machine-to-machine (M2M) standardization protocol with a weight value of 0.0508 has the first rank, followed by adaptation to natural energy sources with a weight value of 0.0479, information security and privacy protection with a weight value of 0.0469, and internet protocol version-6 (IPv6) for low-end devices with weight 0.0467. To validate the proposed method, sensitivity analysis and comparison using existing methods have been conducted.

Low Memory Access Video Stabilization for Low-Cost Camera SoC.

Video stabilization is one of the most important features in consumer cameras. Even simple video stabilization algorithms may need to access the frames several times to generate a stabilized output image, which places a significant burden on the camera hardware. This high-memory-access requirement makes it difficult to implement video stabilization in real time on low-cost camera SoC. Reduction of the memory usage is a critical issue in camera hardware. This paper presents a structure and layout method to efficiently implement video stabilization for low-end hardware devices in terms of shared memory access amount. The proposed method places sub-components of video stabilization in a parasitic form in other processing blocks, and the sub-components reuse data read from other processing blocks without directly accessing data in the shared memory. Although the proposed method is not superior to the state-of-the-art methods applied in post-processing in terms of video quality, it provides sufficient performance to lower the cost of camera hardware for the development of real-time devices. According to my analysis, the proposed one reduces the memory access amount by 21.1 times compared to the straightforward method.

An approach towards improvement of contiguous memory allocation linux kernel: a review

<p><span>The demand of contiguous memory allocation has been expanded in day-to-day life in all the devices. It is achieved in existing systems by using various reservation techniques. There are various other methods to achieve the goal of contiguous memory allocation in linux kernel such as, input output memory management units (IOMMU’s), scatter/gather direct memory access (DMA) and reserved static memory at boot time. But these solutions have its own drawbacks such as, IOMMU requires hardware. However, the configuration of additional hardware's increases the cost. The power consumption of the system and the reserved static memory in the system goes waste when not in used for specific purpose. It is very difficult to access contiguous memory in low-end devices that are unable to provide real contiguous memory. There is one existing method called contiguous memory allocator (CMA), which provides dynamic contiguous memory. It overcomes most of the problems but CMA itself has some drawbacks, which do not provide the guarantee of failure in future of contiguous memory. The motivation behind this study is to review existing contiguous memory allocation (CMA) method by identifying and removing its drawbacks.</span></p>

Real-time object detector for low-end devices

YOLOv4-tiny is a lightweight network designed for low-end devices. It proposes a global model scaling technology, which uses the same scaling method with few convolution filters in each stage of the network, resulting in a small receptive field and low accuracy. To solve this problem, we use different scaling techniques in shallow and deep network stages. For the shallow network stage, the Simple-StemBlock scaling module is proposed to simplify network based on factors such as FLOPs and network fragmentation. The module can effectively reduce computation and improve the diversity of features. For the deep network stage, we consider depth of network and hardware resource constraints, the Depth-CSPBlock scaling module is designed to expand receptive field while keeping the computation as low as possible and layered residual connection is built in the module to enrich semantic information. Besides, mish activation function is adopted in the backbone for higher accuracy. The experimental results show that the accuracy of the proposed method achieves 23.2% AP on MSCOCO test-dev and 66.73 % AP50 on VOC datasets, compared with YOLOv4-tiny, the AP and AP50 increased by 3.3% and 3.7%, respectively. The speed of the proposed method on Jetson TX2 can reach 45.6 frames per second, which is 20% faster than YOLOv4-tiny.

Deep neural network compression by Tucker decomposition with nonlinear response

Deep neural networks have shown impressive performance in many areas, including computer vision and natural language processing. Millions of parameters in deep neural network limit its deployment in low-end devices due to intensive memory and expensive computational cost. In the literature, several network compression techniques based on tensor decompositions have been proposed to compress deep neural networks. Existing techniques are designed in each network unit by approximating linear response or kernel tensor using various tensor decomposition methods. What is more, research has shown that there exists significant redundancy between different filters and feature channels of kernel tensor in each convolution layer. In this paper, we propose a new algorithm to compress deep neural network by considering both nonlinear response and the multilinear low-rank constraint in the kernel tensor. To overcome the resulted difficulty of nonconvex optimization, we propose a convex relaxation scheme such that it can be solved by alternating direction method of multipliers (ADMM) directly. Thus, the Tucker-2 rank and the feature matrix of Tucker decomposition can be determined simultaneously. The effectiveness of the proposed method is evaluated on CIFAR-10 and large-scale ILSVRC12 datasets for CNNs including ResNet-18, AlexNet and GoogleNet. According to our numerical computation, the proposed method is able to obtain highly reduction in model size with a small loss in accuracy. The compression performance of the proposed method is better than existing methods.

Less is More: Lighter and Faster Deep Neural Architecture for Tomato Leaf Disease Classification

To ensure global food security and the overall profit of stakeholders, the importance of correctly detecting and classifying plant diseases is paramount. In this connection, the emergence of deep learning-based image classification has introduced a substantial number of solutions. However, the applicability of these solutions in low-end devices requires fast, accurate, and computationally inexpensive systems. This work proposes a lightweight transfer learning-based approach for detecting diseases from tomato leaves. It utilizes an effective preprocessing method to enhance the leaf images with illumination correction for improved classification. Our system extracts features using a combined model consisting of a pretrained MobileNetV2 architecture and a classifier network for effective prediction. Traditional augmentation approaches are replaced by runtime augmentation to avoid data leakage and address the class imbalance issue. Evaluation on tomato leaf images from the PlantVillage dataset shows that the proposed architecture achieves 99.30% accuracy with a model size of 9.60MB and 4.87M floating-point operations, making it a suitable choice for low-end devices. Our codes and models will be made available upon publication.

Privacy-Preserving Fog Aggregation of Smart Grid Data Using Dynamic Differentially-Private Data Perturbation

The edge of the smart grid has a massive number of power and resource-constrained interconnected devices. Mainly, smart meters report power consumption data from consumer homes, industrial buildings, and other connected infrastructures. Multiple approaches were proposed in the literature to preserve the privacy of consumers by altering the data via additive noise, masking, or other data obfuscation techniques. A significant body of work in the literature employs differential privacy methods with constraining predefined parameters to achieve the optimal trade-off between privacy and utility of the data. However, billing accuracy can be degraded by using such additive noise techniques. We propose a differentially-private model that perturbs data by adding noise obtained from a virtual chargeable battery, while maintaining billing accuracy. Our model utilizes fog-computing data aggregation with lightweight cryptographic primitives to ensure the authenticity and confidentiality of data generated by low-end devices. We describe our differentially-private model with flexible constraints and a dynamic window algorithm to maintain the privacy-budget loss in infinitely generated time-series data. Our experimental results show a possible decrease in data perturbation error by 51.7% and 61.2% for smart meters and fog-computing data aggregators perturbed data, respectively, compared to the commonly used Gaussian mechanism.

A Novel Learning Strategy for the Trade-Off Between Accuracy and Computational Cost: A Touch Modalities Classification Case Study

Wearable systems require resource-constrained embedded devices for the elaboration of the sensed data. These devices have to host energy-efficient artificial intelligence (AI) algorithms to output information to a human user. In this regard, the single-layer feed-forward neural networks (SLFNNs) proved to be very effective for deployment on very low-end devices. SLFNNs showed promising results when the goal was to balance the trade-off between accuracy and predictor complexity. Nevertheless, the elaboration system, which consists of data pre-processing and features extraction stages, affects the resource occupancy, computational cost, and also eventual generalization performance. In this paper, we propose a learning strategy based on the out-of-samples technique that leads to finding the configurations of the elaboration unit and the parameters of the predictor that best balance the generalization accuracy and the computational cost of the whole system. An intensive study over the touch modalities classification problem shows that the best solution in terms of the only accuracy is achieved by the SLFNNs, outperforming existing algorithms. Moreover, the learning procedure selects the configuration of the processing and features extraction stages, and the parameters of the predictor that present the best trade-off between generalization performance and computational cost.

PocketNet: Extreme Lightweight Face Recognition Network Using Neural Architecture Search and Multistep Knowledge Distillation

Deep neural networks have rapidly become the mainstream method for face recognition (FR). However, this limits the deployment of such models that contain an extremely large number of parameters to embedded and low-end devices. In this work, we present an extremely lightweight and accurate FR solution, namely PocketNet. We utilize neural architecture search to develop a new family of lightweight face-specific architectures. We additionally propose a novel training paradigm based on knowledge distillation (KD), the multi-step KD, where the knowledge is distilled from the teacher model to the student model at different stages of the training maturity. We conduct a detailed ablation study proving both, the sanity of using NAS for the specific task of FR rather than general object classification, and the benefits of our proposed multi-step KD. We present an extensive experimental evaluation and comparisons with the state-of-the-art (SOTA) compact FR models on nine different benchmarks including large-scale evaluation benchmarks such as IJB-C and MegaFace. PocketNets have consistently advanced the SOTA FR performance on nine mainstream benchmarks when considering the same level of model compactness. With 0.92M parameters, our smallest network PocketNetS-128 achieved very competitive results to recent SOTA compacted models that contain up to 4M parameters. Training codes and pre-trained models are public. <xref ref-type="fn" rid="fn1" xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">¹</xref>

Research on AI (Artificial Intelligence) with the Robotic Arm

Robot operating system (ROS) is robot middleware. Although ROS is not an operating system, it provides the necessary services designed for a heterogeneous cluster of servers, controlling low-end devices, deploying functions, transmitting messages, and managing low-end devices. A set of ROS-based processes represented as architectural graphs that can receive and concatenate sensor data, state control, scheduling, actuators, and other messages. The application of ROS to the product is in order to test and simulate the operation of the robot arm, thereby supporting researchers, improving, developing and deploying applications on robots smoothly. Stemming from the above analysis, the topic completely meets the urgency posed by modern society.

Brain tumor magnetic resonance images classification based machine learning paradigms

Cancer of the nervous system is one of the most common types of cancer in the world and mostly due to presence of a tumour in the brain. The symptoms and severity of the brain tumour depend on its location. The tumour within the brain may develop from nerves, dura (meningioma), pituitary gland (pituitary adenoma), or from the brain tissue itself (glioma). In this study we proposed a feature engineering approach for classification magnetic resonance imaging (MRI) of 3 kinds of most common brain tumour, i.e. glioma, meningioma, pituitary, and no-tumour. Here 5 machine learning classifiers were used, i.e. support vector machine, K-nearest neighbour (KNN), Naive Bayes, Decision Tree, and Ensemble classifier with their paradigms. The handcrafted features such as histogram of oriented gradients, local binary pattern features, and grey level co-occurrence matrix are extracted from the MRI, and the feature fusion technique is adopted to enhance the dimension of feature vector. The Fine KNN outperforms among the classifiers for recognition of 4 kinds of MRI: glioma, meningioma, pituitary, and no tumour, and achieved 91.1% accuracy and 0.95 area under the curve (AUC). The proposed method, i.e. Fine KNN, achieved 91.1% accuracy and 0.96 AUC. Furthermore, this model has the possibility to integrate in low-end devices unlike deep learning, which required a complex system.

A systematic analysis of identity based encryption (IBE)

Light-weight cryptography is a major research area due to the minimization of the size of the devices utilized for such services. The associated security threats do increase as their applications are more now. Identity-Based Encryption (IBE) with its wide range of cryptographic schemes and protocols is specifically found suitable for low-end devices that have much resource constraint. This work describes various schemes and protocols in IBE. In this paper an analysis of IBE schemes and the various attacks they are prone to are discussed. The future trends are found to be very promising and challenging.

Mobile Augmented Reality for Low-End Devices Based on Planar Surface Recognition and Optimized Vertex Data Rendering

Mobile Augmented Reality (MAR) is designed to keep pace with high-end mobile computing and their powerful sensors. This evolution excludes users with low-end devices and network constraints. This article presents ModAR, a hybrid Android prototype that expands the MAR experience to the aforementioned target group. It combines feature-based image matching and pose estimation with fast rendering of 3D textured models. Planar objects of the real environment are used as pattern images for overlaying users’ meshes or the app’s default ones. Since ModAR is based on the OpenCV C++ library at Android NDK and OpenGL ES 2.0 graphics API, there are no dependencies on additional software, operating system version or model-specific hardware. The developed 3D graphics engine implements optimized vertex-data rendering with a combination of data grouping, synchronization, sub-texture compression and instancing for limited CPU/GPU resources and a single-threaded approach. It achieves up to 3× speed-up compared to standard index rendering, and AR overlay of a 50 K vertices 3D model in less than 30 s. Several deployment scenarios on pose estimation demonstrate that the oriented FAST detector with an upper threshold of features per frame combined with the ORB descriptor yield best results in terms of robustness and efficiency, achieving a 90% reduction of image matching time compared to the time required by the AGAST detector and the BRISK descriptor, corresponding to pattern recognition accuracy of above 90% for a wide range of scale changes, regardless of any in-plane rotations and partial occlusions of the pattern.