The aim of this study is to give elements to predict time performances and power consumption performances in CNN implementation on embedded CPU boards. CNNs are effective in classifying EEG signals but are computationally intensive, which leads to increased energy consumption. To achieve this objective, we compare the implementation of a trained convolutional neural network (CNN) for epilepsy detection, analyzing EEG signals on some resource-constrained embedded platforms commonly used in IoT applications. We evaluate the ease of deployment and the time and energy performance metrics across four different CPU embedded boards to guide and anticipate future implementations of CNNs on embedded solutions. Depending on the board, (ARM based, 32 bits / 64 bits architecture / Multicore) the execution time performances vary from 10 time more to 10 time less compare to a standard Laptop. Obviously, the power consumption varies accordingly. Multicore platforms require also specific tools and we evaluate the benefits of it compared to a standard approach.

With the growing use of biomedical signal acquisition systems such as EEG, ensuring high-quality data acquisition in resource-constrained environments remains a critical challenge. This study evaluates how data quality-specifically noise, error, and loss-affects the performance of a CNN-based epileptic seizure detection system in terms of accuracy, execution time, and power consumption. We simulate different corruption scenarios and propose strategies for tolerance assessment and lightweight compensation. The results provide insights for building adaptive EEG-based biomedical systems.

This study explores several machine learning methodologies to effectively classify distinct hand gestures using data collected by a gauntlet embedded with multiple surface electromyography (sEMG) sensors placed to capture muscle activation patterns. The analysis includes correlations between sensor data, filtering techniques, and a variety of machine learning algorithms. These efforts had the goal of identifying approaches that could facilitate the development of a commercially viable device. By evaluating the classification accuracy and computational efficiency of each method during the validation phase, this study seeks to identify techniques to be used for further refinement in this kind of product.

This paper intends to use a small protocol to detect stroke disease on a patient by using signals provided by only three EEG probes. To achieve this objective, we compare the performances in terms of accuracy and time of six machine learning (ML) algorithms (Random Forest, Logistic Regression, Support Vector Machine, K-Nearest Neighbor, Decision Tree and CatBoost) during a process of EEG-based classification pathology. We use a database of EEG recording signals collected by three electrodes, established by Beijing University of Chinese Medicine and carried out on subjects healthy or affected by strokes when they are exposed to the vision of planes of five different colors. The subjects are known to be healthy or affected by strokes. The records are used to train each algorithm for 70% of the population, and the performances are estimated on the remaining 30%. Then the process is repeated one hundred times when changing the set used for training and the set used to test. We then consider a statistic on the results obtained using each method for comparison. Our results show that the SVM algorithm is the most efficient in terms of the accuracy of the results, and can detect stoke disease with a reliability of 70%.

Ever since the first isolation of graphene, the semimetal has grown appreciable and has been attracting increasing interest. This interest is reinforced by monolayer graphene's remarkable electronic properties and its usage in revolutionary device developments and applications. However, obtaining monolayer graphene which can be deployed for expansion of experiments in 2D physics comes with its own limitations like high human interventions that requires significant experience, is highly time consuming since it involves repetitive tasks and recognizing graphene crystallites from millions of thicker graphite flakes with other undesired particles is strenuous. Here, we report an approach to detect and discriminate monolayer graphene from other alternating layers of graphene and subsisting substrate impurities. We present a region of interest-based image segmentation process to extricate inapplicable information from the image and extract graphene particles. We, then apply an intensity-based detection model leveraging the characteristic color information to differentiate monolayer graphene from other particles and it is observed that the red color space of the monolayer graphene differs 1.8--6%, green 2.5--8% and blue differ 2.5% to 3% from the surrounding background pixels. We also describe an implementation of our algorithm in a semi-automatic system suitable with our needs.

This paper deals with the problem of counting the amount of pollen grains (density of pollen) in a view acquired on a daily basis by a dedicated device. The grains are stuck on a ribbon which is analyzed by a microscope. This task is currently performed by a human operator who has to analyze quite 300 microscopic slides representing 10% of the ribbon. This task requires a particularly high concentration. Moreover, the viewer has to identify the pollen grains among water drops and soot spots. We propose an algorithm composed of four main steps to perform this task automatically. The basic detection of pollen grains, relying on their shape and color, is completed by pre-processing and post-processing operations to handle specific cases like broken grains or grains at the border of images. Finally the efficiency is improved by using a neural network to refine the results. Our automatic counter has been compared with the pollen density obtained by the manual counting and our program has proved its high accuracy.

In this paper we propose a real time (100 frames/sec) Region of Interest (ROI) detection algorithm based on the Integral Image calculation and its implementation on FPGA, while considering algorithm’s optimization and power consumption. This system is designed for an embedded system connected to a CCD sensor inserted on glasses for precise eye tracking purposes. The ROI detection permits to select the useful data for a precise eye tracker algorithm. Compared with the state-of-the-art methods, this architecture proves its efficiency considering the processing speed and the power consumption for data flow operations in real time. We also explore the possibility to handle the computation load by embedded processors in order to show that FPGAs are ten time more efficient when considering power dissipation.

This paper deals with the problem of counting the amount of pollen grains (density of pollen) in a view acquired on a daily basis by a dedicated device. The grains are stuck on a ribbon which is analyzed by a microscope. This task is currently performed by a human operator who has to analyze quite 300 microscopic slides representing 10% of the ribbon. This task requires a particularly high concentration. Moreover, the viewer has to identify the pollen grains among water drops and soot spots. We propose an algorithm composed of four main steps to perform this task automatically. The basic detection of pollen grains, relying on their shape and color, is completed by pre-processing and post-processing operations to handle specific cases like broken grains or grains at the border of images. Finally the efficiency is improved by using a neural network to refine the results. Our automatic counter has been compared with the pollen density obtained by the manual counting and our program has proved its high accuracy.

In this paper, a new preprocessing algorithm to qualify images of different pollen grains for further processing is proposed. This algorithm provides a score related to the sharpness of the image and will be used to automatically adjust the focal length of a microscope that magnifies the image. The obtained score has been compared to four quality metrics generally used to estimate the clarity of an image and to a reference made by a human. The results of the simulations show that the proposed algorithm combines better performance with low complexity on the set of images.

ISEP-R0B0 is a project which combines a small programmable robot and a visual 1 programming language. Its goal is to provide a full-fledged system at a very low cost, 2 targeting schools but also informal learning situations such as after-class activities. 3 Through the programming and observation of the behavior of the robot, students can 4 learn notions either related to Computer Science or Science, Technology, Engineering, 5 and Mathematics. Since ISEP-R0B0 is still at an early stage, this article focuses on 6 introducing the design of the system and two case studies we plan on conducting shortly. 7 Introduction and Context 8 Research in Computer Science Education (CSE) has long tried to introduce robots 9 in programming courses. Oftentimes, the objective is to foster students’ interest and 10 creativity through “the design of tangible and interactive object using programmable 11 hardware” [14], also known as physical computing. In this regard, results indicate 12 that students experience an increase in motivation [7, 17] and that underrepresented 13 populations in Computer Science (CS) courses feel empowered [16]. However, learning 14 outcomes can vary depending on the context and course taught [2, 4]. 15 Two aspects of programming often cause difficulties to beginners [3,9]: the complexity 16 of a programming language and the structural instructions of a programming algorithm. 17 By focusing on the design of algorithms rather than on the code implementation itself, 18 Visual Programming Languages (VPL) can alleviate these two issues. We can mention 19 the cases of Scratch [12] and App Inventor [6] which were both used to teach fundamental 20 CS principles. A more recent example is the BBC micro:bit project, from which the 21 design of ISEP-R0B0 is inspired. The micro:bit is a “pocket-sized, codeable computing 22 device” which is programmable with an online block-based VPL [1]. Though useful as a 23 tool to foster creativity and increase motivation [17], we could not find any publication 24 identifying the effects of using the micro:bit in acquiring CS and programming knowledge. 25 ISEP-R0B0 is composed of a programmable robot and a VPL. We expect learners to 26 program ISEP-R0B0, and to see the robot itself as a playful tool for the embodiment 27 of CS or Science, Technology, Engineering, and Mathematics (STEM) concepts. Other 28 systems using block-based VPLs were previously designed with similar intents: .NET 29 Gadgeteer [7], LEGO Mindstorms [2], Thymio-II [10], and more recently micro:bit [1]. 30 However, distinctions can be drawn on the granularity of each VPL and on the cost 31 of each product. Indeed, ISEP-R0B0 provides low-level instruction blocks of code to 32 program the robot (in opposition to LEGO Mindstorms or .NET Gadgeteer which 33

In this paper we consider the case of an embedded system connected to a CCD sensor inserted on glasses for precise eye tracking purposes. We describe an image compression algorithm and its implementation when considering the algorithm cost and the power consumption. The principle is to transmit a high resolution image while keeping a precise definition of the pupil edge in order to post process pupil localization. We compare performances between processors and FPGAs, and to demonstrate FPGAs allow data flow algorithms to reach the objectives with less power.

Frame format conversion represents a key part of video display systems, as usually the transmission video stream resolution is different from the screen resolution of the digital display. In this paper a contour-preserving approach for video up-scaling is proposed, along with a low-cost, efficient FPGA architecture for one of the most used format conversions, namely PAL to XGA. The algorithm first checks for edges existence within the region around the pixel to be interpolated and adapts the scaling method following the contour direction. This adaptive system keeps the edges sharp and provides good quality scaled video sequences when compared with other scaling methods. Moreover, we propose an FPGA implementation for handling the algorithm reload, very efficient in terms of resources and complexity.

In this paper, we describe a novel technique for the extraction of object shapes from Terahertz images using a 3D graph-cut segmentation scheme. This approach to segmentation includes images in temporal domain by creating nodes and edges between consecutive images in order to obtain improved segmentation results and compensate for the high levels of noise in the Terahertz images. The foreground and background pixels are defined to provide hard constraints for segmentation. Graph cuts are thus used to extract the object shapes in absence of any other direct mean of object identification.

This article describes an experiment conducted during six years at the Institut Supérieur d'Electronique de Paris based on a Project Oriented Learning approach to teach the fundamentals of Electronics. The proposed teaching framework has a dynamic structure, as it adapts and modifies the terms of annual assessments to foster the motivation and interest of students. Results show the effectiveness and the value of this approach for student motivation.

Biomedical and health applications are representing nowadays a very attractive area for electronics devices. The state of the art shows a multitude of solutions based on in body sensors devices to measure and survey human physiological parameters enabling a distance medical monitoring systems and higher out-of-hospital care services for patients. These emergent systems have a lot of challenges related to their power consumption, size and complexity. In this article, we focus on one of the technological locks of the electronic gastric pill, which is the antenna. A miniaturized patch antenna a key element of the receiving system is presented. The antenna size and performances are very important to consider as it have to be integrated on a jacket in order to collect the transmitted information from an electronic gastric pill to trace cartography of the collected data. Different miniaturization techniques of patch antennas are combined with the use high dielectric constant substrate in order to reach a tradeoff between the antenna performances and the size. The designed antenna resonates in the desired ISM frequency band (430 MHz) and was simulated and optimized with the presence of a model of the human body for a more realistic use case. The antenna size reduction achieves 81.56% with good performances in terms of gain (-29.4dB).

The reliability of new SRAM FPGA (Field Programmable Gate Array) devices, which are the first components launched for each new generation of transistor, is difficult to estimate. Their increasing use on electronic boards in both terrestrial and space applications necessitates the development of fault tolerant techniques in the wake of growing soft error rates (SER). In this article, In this article, a concurrent error detection and correction scheme using residue codes is proposed and designed. The results express the gain for power consumption and circuit area compared for other solutions for fault detection or fault correction.

In this work, the impact of different power reduction techniques at architectural level on modern FPGAs is evalua ted. Starting from the 65 nm technology node, manufacturers have implemented a number of technologies to reduce overall powe r consumption. In our study we are performing real life measurements and proposing several architecture level power reduc tion techniques. Our solutions are then implemented and evaluat ed in an image processing intellectual property block. Obtainedresults demonstrate that certain techniques such as clock gating ma y result in up to 90% power savings.