GB RAM Research Articles

Digitally improving UK police surveillance and incidence response using real-time crowd reporting app: Digipolice

Overtime, evolving technologies has made ways of monitoring and controlling crime and action in the United Kingdom (UK) become more challenging which needs optimization with the help of technologies. Also with the rate of crime across cities in the UK like mass shootings, murder and gang gun battles, the UK Police system within localities are overwhelmed or may not efficiently cope with crime alerts and incidences coming in at increasing frequencies. However looking in the area of real time crowd reporting, there is need to leverage technology to enhance public safety, improve incident response, and foster community engagement. This necessitated the aim of this study to develop a digital policing surveillance app for crowd reporting in the UK. By utilizing digital policing surveillance apps, the aim is to improve crowd management capabilities by enabling real-time reporting of incidents and observations from both the public and law enforcement personnel. This was positioned to facilitate better situational awareness, efficient resource allocation, and effective response strategies. To achieve the objectives, Python Programming Language 3.8 were deployed together with a Django Framework and a Microsoft Visual Studio Build Tools. In the DP system, the Model-View-Controller (MVC) pattern, which divides an application into three interconnected components—the model, the view, and the controller—was used for the architectural design of the system. For the database, Django's ORM makes it simple to support many databases, and in this project straightforward SQLite3 database were utilised which is ideal for small-footprint applications. Using WebSockets to establish real-time, bidirectional communication between web clients and servers, the user interface and intelligence board for the system were developed. A clean designed frontend design system with Bootstrap version 5—an open-source, reusable design framework created by Twitter—was selected for its easy user interface. The system was put through 50 concurrent load tests with actual users swarming the app since the DP app was subjected to hundreds of people complaining in real time. 2. The host computer had a Core i5 vPro CPU running at 4GHz and 4GB of RAM. The result shows that 50 users successfully generated an average of 29.7 requests per second. The system scaled up nicely to 50 users and reached its highest response time of 2600 milliseconds (2.6 seconds) in total. A dramatic rise in activity and a longer response time occurred after there were more than 50 users. The system could not operate at its peak efficiency with 100 users because response times were inconsistent and exceeded 7000 milliseconds (7 seconds) for each request. The effectiveness of the online deployment was evaluated with respectable results, and the system requirements also underwent successful testing. The appropriate investigation of ethical issues was done to meet requirements. DigiPolice has the potential to develop into a cutting-edge programme for real-time crowd surveillance and incident reporting. UK Police can employ it to as advancement to its “My Police” with as fast-track real-time and collaborative crowd reporting enhancements.

EpiCarousel: memory- and time-efficient identification of metacells for atlas-level single-cell chromatin accessibility data.

Recent technical advancements in single-cell chromatin accessibility sequencing (scCAS) have brought new insights to the characterization of epigenetic heterogeneity. As single-cell genomics experiments scale up to hundreds of thousands of cells, the demand for computational resources for downstream analysis grows intractably large and exceeds the capabilities of most researchers. Here, we propose EpiCarousel, a tailored Python package based on lazy loading, parallel processing, and community detection for memory- and time-efficient identification of metacells, i.e. the emergence of homogenous cells, in large-scale scCAS data. Through comprehensive experiments on five datasets of various protocols, sample sizes, dimensions, number of cell types, and degrees of cell-type imbalance, EpiCarousel outperformed baseline methods in systematic evaluation of memory usage, computational time, and multiple downstream analyses including cell type identification. Moreover, EpiCarousel executes preprocessing and downstream cell clustering on the atlas-level dataset with 707043 cells and 1154611 peaks within 2 h consuming <75 GB of RAM and provides superior performance for characterizing cell heterogeneity than state-of-the-art methods. The EpiCarousel software is well-documented and freely available at https://github.com/biox-nku/epicarousel. It can be seamlessly interoperated with extensive scCAS analysis toolkits.

A chaotic based image encryption scheme using elliptic curve cryptography and genetic algorithm

In the era of digital communication and data security, image encryption plays a crucial role in safeguarding sensitive information. Protecting sensitive visual data from unauthorized access drives the pursuit of advanced image encryption methods. This paper proposes a novel approach to enhance image encryption by combining the power of a chaotic map, elliptic curve cryptography, and genetic algorithm. The chaotic map, specifically Arnold’s cat map, is employed to introduce chaos and randomness into the encryption process. The proposed image encryption process involves applying Arnold’s cat map for shuffling the pixel positions, followed by elliptic curve cryptography for encrypting the pixel values using public and private keys. Additionally, a genetic algorithm is employed to optimize the key generation process, enhancing the security of the encryption scheme. The combined utilization of these techniques aims to achieve a high level of confidentiality and robustness in image encryption. The algorithm underwent thorough analysis. It achieved a maximum entropy score of 7.99, indicating a high level of randomness and unpredictability in the encrypted data. Additionally, it exhibited near-zero correlation, which suggests strong resistance against statistical attacks. Moreover, the cryptographic range of possible keys was found to be 2511\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$2^{511}$$\\end{document}. This extensive key space makes the algorithm highly resilient against brute force attacks. It took only 0.5634 s to encrypt a moderately sized 512×512\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$512\ imes 512$$\\end{document} pixel image with an 8-bit image on a standard desktop computer with a 2.3 GHz processor and 16 GB of RAM. The experimental findings confirm that the proposed approach is highly effective and efficient in safeguarding sensitive image data from unauthorized access and potential attacks. This scheme has the benefit of allowing us to protect our private image data while it’s being transmitted.

Biomechanical Effects of Different Auxiliary–Aligner Designs on the Rotation of an Upper Canine: A Finite Element Analysis of a Specific Patient

Aim: The objective of this research has been to apply a specific simulation to a patient to assess the biomechanical consequences of rotating an upper canine tooth through different attachment–aligner configurations and to predict the most efficient design using a three-dimensional finite element model of a full maxillary arch of a specific patient. Materials and methods: This was obtained by combining Cone-Beam Computed Tomography (CBCT) with the aim of reconstructing tooth roots and bone tissues, and Surface Structured-Light Scanning for creating digital tooth crown models from the patient’s impressions. This model was imported into the finite element solver (Ansys® 17). Three different attachment–aligner combinations were created through the exploitation of computer-aided design (CAD) procedures, i.e., without attachments, with a couple of attachments and with an attachment and a pressure point. For each simulation, the resulting force–moment (MF) system applied by the aligner to the target tooth, as well as the tooth displacement and rotation, was computed using a workstation based on Intel Xeon CPU E3-1245 v3@3.40 GHz and 16 GB RAM. Simulations reported that by adding the pressure point and the attachment to the standard aligner the amount of Moment z (Mz) delivered to the tooth increased almost two times. Results and conclusions: The maximum tooth displacement (0.85 mm) was obtained with the attachment and pressure point aligner, while the lowest (0.058 mm) was obtained with use of a couple of attachments. Both the attachment and the pressure point have the potential to enhance the appliance’s effectiveness. Particularly, the pressure point showed a higher influence on the load absolute value. The method applied in the present study should be used to retrieve the best design configuration for each patient and specific tooth movement.

Trapping Sets Search Using the Method of Mixed Integer Linear Programming with a Priori List of Variable Nodes

Purpose of research is to develop a new high-speed method for searching trappin sets in graph codes, ensuring the completeness of the search.Methods. There are two approaches to finding trappin sets. The first, based on the Monte Carlo method with a biased probability estimation using Importance Sampling, involves the use of a decoder. The advantage of this approach is its high performance. The disadvantages are the dependence on decoder parameters and channel characteristics and the finite probability of missing trappin sets. The second approach is based on the use of linear programming methods. The advantage of this approach is the completeness of the resulting list of trappin sets, due to its independence from the decoder parameters and channel characteristics. The disadvantage of this approach is its high computational complexity. In the article, within the framework of the second approach, a new method for searching trappin sets with less computational complexity is proposed. The method involves solving a mixed integer linear programming problem using an a priori list of code vertices participating in the shortest cycles in the code graph. Results. Using the proposed method, a search for trappin sets was performed for several low-density codes. For this purpose, the mathematical linear programming package IBM CPLEX version 12.8 was used, which was run on 32 threads of a 16-core AMD Ryzen 3950X processor with 32GB of RAM (DDR4). In the Margulis code (2640, 1320), using the proposed method, the trappin set TS(6,6) was found in a time of 0.53 s. The speedup provided by the method proposed in the paper compared to the Velazquez-Subramani method is 8252.415 times. Thanks to the high speed and completeness of the search, trappin sets were found for the first time TS(62,16) and TS(52,14) in the Margulis code (4896, 2474 ).Conclusion. The paper proposes a new method for searching trapping sets by solving a mixed integer linear programming problem with an a priori list of code. The method is fast and provides completeness of the search.

Predicting the ages of galaxies with an artificial neural network

ABSTRACT We present a new method of predicting the ages of galaxies using a machine learning (ML) algorithm with the goal of providing an alternative to traditional methods. We aim to match the ability of traditional models to predict the ages of galaxies by training an artificial neural network (ANN) to recognize the relationships between the equivalent widths of spectral indices and the mass-weighted ages of galaxies estimated by the magphys model in data release 3 (DR3) of the Galaxy and Mass Assembly (GAMA) survey. We discuss the optimization of our hyperparameters extensively and investigate the application of a custom loss function to reduce the influence of errors in our input data. To quantify the quality of our predictions we calculate the mean squared error (MSE), mean absolute error (MAE) and R2 score for which we find MSE = 0.020, MAE = 0.108 and R2 = 0.530. We find our predicted ages have a similar distribution with standard deviation σp = 0.182 compared with the GAMA true ages σt = 0.207. This is achieved in approximately 23 s to train our ANN on an 11th Gen Intel Core i9-11900H running at 2.50 GHz using 32 GB of RAM. We report our results for when light-weighted ages are used to train the ANN, which improves the accuracy of the predictions. Finally, we detail an evaluation of our method relating to physical properties and compare with other ML techniques to encourage future applications of ML techniques in astronomy.

Driver Drowsiness Detection using Machine Learning Approach

Driver fatigue is a major cause of traffic accidents. The number of deaths and injuries increases every year around the world. Traffic accidents can be reduced by detecting driver fatigue. This article describes machine learning for sleep detection. Face detection is used to detect the driver's eye area and use this as a reference for eye tracking in subsequent frames. Finally, visual images are used to detect sleep and a warning system is created. This method is divided into three stages: face detection, eye detection, and fatigue detection. Image processing is used to recognize the driver's face and then extract the image of the driver's eyes to detect fatigue. HAAR face detection algorithm outputs the image and then adjusts face detection based on the output. CHT is then used to track the eyes of the visible face. Check the eyes using EAR (Early Evaluation). The proposed system was tested using the proposed system on a Raspberry pi 3 Model B with 1 GB RAM using Logitech HD Webcam C270. According to some video tests, average eye contact and tracking accuracy can reach 95.0%. Therefore, it is a cheaper and better solution for a tired driver to ask to find the road immediately.

Clinical validity and precision of deep learning-based cone-beam computed tomography automatic landmarking algorithm.

This study was performed to assess the clinical validity and accuracy of a deep learning-based automatic landmarking algorithm for cone-beam computed tomography (CBCT). Three-dimensional (3D) CBCT head measurements obtained through manual and automatic landmarking were compared. A total of 80 CBCT scans were divided into 3 groups: non-surgical (39 cases); surgical without hardware, namely surgical plates and mini-screws (9 cases); and surgical with hardware (32 cases). Each CBCT scan was analyzed to obtain 53 measurements, comprising 27 lengths, 21 angles, and 5 ratios, which were determined based on 65 landmarks identified using either a manual or a 3D automatic landmark detection method. In comparing measurement values derived from manual and artificial intelligence landmarking, 6 items displayed significant differences: R U6CP-L U6CP, R L3CP-L L3CP, S-N, Or_R-R U3CP, L1L to Me-GoL, and GoR-Gn/S-N (P<0.05). Of the 3 groups, the surgical scans without hardware exhibited the lowest error, reflecting the smallest difference in measurements between human- and artificial intelligence-based landmarking. The time required to identify 65 landmarks was approximately 40-60 minutes per CBCT volume when done manually, compared to 10.9 seconds for the artificial intelligence method (PC specifications: GeForce 2080Ti, 64GB RAM, and an Intel i7 CPU at 3.6 GHz). Measurements obtained with a deep learning-based CBCT automatic landmarking algorithm were similar in accuracy to values derived from manually determined points. By decreasing the time required to calculate these measurements, the efficiency of diagnosis and treatment may be improved.

Development of an IoT-Based Drowning Detection System for Private Swimming Pools

Drowning is among the top three leading causes of unintentional death worldwide, accounting for over 8% of all injury-related deaths. Annually, over 22,000 deaths occur due to drowning in adults, teenagers, and children. The highest rate is among children aged 0–15 years. Available solutions to this murder (drowning) involve traditional fencing of swimming pools, the use of wrist bands with RF connections, heartbeat sensors, monitoring systems using LASER and LDR, etc. These methods are not sufficient to solve the problems, as past methods involve wearable buildings systems. This paper proposes a method that is dependent on the swimming pool (the swimmer does not need to wear any device). The proposed system utilizes multiple sensor technology and IoT technology; PIR sensor for motion detection and Raspberry Pi module v3 for capturing and detecting possible drowning events via a trained model using yolo v5 architecture integrated into Raspberry Pi 4 8GB RAM. The system sends an email message to the authorized user if it senses motion around the swimming pool environment, detects a human in an un-save zone, and sends an alert of possible drowning.

Monte Carlo Simulation-Based Regression Tree Algorithm for Predicting Energy Consumption from Scarce Dataset

Most data-driven techniques rely on the availability of data. Hence, when the data provided is not sufficient, the algorithm might not work as intended. Thus, it is important to be able to predict the dynamics of the data, even when the number of available data is low, or scarce. This study aimed to predict the power consumption of a building given a scarce dataset via a novel Monte Carlo simulation-based Regression Tree (MCRT) algorithm. The main idea is to train Monte Carlo simulation on each leaf generated by the regression tree algorithm. Thus, the prediction no longer depends on the average of the samples contained in the leaf, but now depends on the probability of the samples. The proposed algorithm was validated on 2 datasets obtained from Universitas Widya Dharma Pontianak (UWDP), Indonesia and Trapeznikov Institute of Control Sciences (TICS), Russia. To show that the MCRT algorithm is better than the RT algorithm a two-tail hypothesis was proposed. Based on the experiments which were run on Python software with 16 GB RAM, 7th Gen Core i7 machine on 50 datasets randomly generated from the UWDP electrical data, it can be concluded that the MCRT algorithm performs better than the previous RT algorithm used to model scarce datasets with P-value = 0.000319. Furthermore, the proposed algorithm improves the model predictive accuracy of the RT algorithm by up to 2%.

ImageNet classification with Raspberry Pis: federated learning algorithms of local classifiers

PurposeThis study aims to propose novel edge device-tailored federated learning algorithms of local classifiers (stochastic gradient descent, support vector machines), namely, FL-lSGD and FL-lSVM. These algorithms are designed to address the challenge of large-scale ImageNet classification.Design/methodology/approachThe authors’ FL-lSGD and FL-lSVM trains in a parallel and incremental manner to build an ensemble local classifier on Raspberry Pis without requiring data exchange. The algorithms load small data blocks of the local training subset stored on the Raspberry Pi sequentially to train the local classifiers. The data block is split into k partitions using the k-means algorithm, and models are trained in parallel on each data partition to enable local data classification.FindingsEmpirical test results on the ImageNet data set show that the authors’ FL-lSGD and FL-lSVM algorithms with 4 Raspberry Pis (Quad core Cortex-A72, ARM v8, 64-bit SoC @ 1.5GHz, 4GB RAM) are faster than the state-of-the-art LIBLINEAR algorithm run on a PC (Intel(R) Core i7-4790 CPU, 3.6 GHz, 4 cores, 32GB RAM).Originality/valueEfficiently addressing the challenge of large-scale ImageNet classification, the authors’ novel federated learning algorithms of local classifiers have been tailored to work on the Raspberry Pi. These algorithms can handle 1,281,167 images and 1,000 classes effectively.

A Biobanking System for Diagnostic Images: Architecture Development, COVID-19-Related Use Cases, and Performance Evaluation.

Systems capable of automating and enhancing the management of research and clinical data represent a significant contribution of information and communication technologies to health care. A recent advancement is the development of imaging biobanks, which are now enabling the collection and storage of diagnostic images, clinical reports, and demographic data to allow researchers identify associations between lifestyle and genetic factors and imaging-derived phenotypes. The aim of this study was to design and evaluate the system performance of a network for an operating biobank of diagnostic images, the Bio Check Up Srl (BCU) Imaging Biobank, based on the Extensible Neuroimaging Archive Toolkit open-source platform. Three usage cases were designed focusing on evaluation of the memory and computing consumption during imaging collections upload and during interactions between two kinds of users (researchers and radiologists) who inspect chest computed tomography scans of a COVID-19 cohort. The experiments considered three network setups: (1) a local area network, (2) virtual private network, and (3) wide area network. The experimental setup recorded the activity of a human user interacting with the biobank system, which was continuously replayed multiple times. Several metrics were extracted from network traffic traces and server logs captured during the activity replay. Regarding the diagnostic data transfer, two types of containers were considered: the Web and the Database containers. The Web appeared to be the more memory-hungry container with a higher computational load (average 2.7 GB of RAM) compared to that of the database. With respect to user access, both users demonstrated the same network performance level, although higher resource consumption was registered for two different actions: DOWNLOAD & LOGOUT (100%) for the researcher and OPEN VIEWER (20%-50%) for the radiologist. This analysis shows that the current setup of BCU Imaging Biobank is well provisioned for satisfying the planned number of concurrent users. More importantly, this study further highlights and quantifies the resource demands of specific user actions, providing a guideline for planning, setting up, and using an image biobanking system.

融合B2C和G2C的Metaverse： 旅游业的新商业模式

The Metaverse, a buzzword after last pandemic, is accepted as a functional tool enable for multi sided platforms in a single phygital world. The key contributions of this article include an organized evaluation of metaverse concepts, components, main facilitator technologies and implementation in tourism and a critical analysis of how metaverse can be facilitated in tourism including designing a new metaverse ecosystem by means of Metaverse as a service (MaaS) platform. In designed metaverse, it is aimed to merge two business models B2C &amp;amp;G2C to construct a new hybrid business model B2CG2Ccamping, glamping activities. In order to concrete the designed metaverse, two scenarios were generated by means of scenario-based approach. The metaverse designed by means of Spatial.io on a desktop computer with a 512 GB capacity with processor 11th Gen Intel(R) Core (TM) i9-11950H @ 2.60GHz 2.61 GHz, 16.0 GB RAM, 64-bit -Windows 10 Enterprises. Results show that designed metaverse is applicable for tourism activities such as glamping as a case on this study. Besides, the designed ecosystem approved the two business models can be merged in a one platform: Metaverse.

Online thermal field prediction for metal additive manufacturing of thin walls

Various data-driven modeling methods have been developed to predict the thermal field in metal additive manufacturing (AM). The generalization capability of these models has been shown with simulation, but rarely tested with online physical printing. Instead, this paper aims to study a practical issue in metal AM, i.e., how to predict the thermal field of yet-to-print parts online when only a few sensors are available. This work proposes an online thermal field prediction method using mapping and reconstruction, which could be integrated into a metal AM process for online performance control. Based on the similarity of temperature curves (curve segments of a temperature profile of one point), the thermal field mapping applies an artificial neural network to estimate the temperature curves of points on the yet-to-print layer from measured temperatures of certain points on the previously printed layer. With measured/predicted temperature profiles of several points on the same layer, the thermal field reconstruction proposes a reduced order model (ROM) to construct the temperature profiles of all points on the same layer, which could be used to build the temperature field of the entire layer. The training of ROM is performed with an extreme learning machine (ELM) for computational efficiency. Fifteen wire arc AM experiments and nine simulations are designed for thin walls with a fixed length and unidirectional printing of each layer. The test results indicate that the proposed prediction method could construct the thermal field of a yet-to-print layer within 0.1 s on a low-cost desktop computer (Intel Core i7-3770 CPU @ 3.40GHz processor, 24.0 GB RAM). Meanwhile, the method has acceptable generalization capability in most cases from lower layers to higher layers in the same simulation, as well as from one simulation to a new simulation on different AM process parameters. More importantly, after fine-tuning the proposed method with limited experimental data, the relative errors of all predicted temperature profiles on a new experiment are sufficiently small, which demonstrates the applicability and generalization of the proposed thermal field prediction method in online applications for metal AM.

Poseidon - a Collaborative Environment for the Daratumumab Sensitivity Research Trial

Precision Oncology Software Environment Interoperable Data Ontologies Network (POSEIDON) was conceived at City of Hope (COH) as a cloud-based computational and data storage platform. The platform brings together patient data from the COH electronic medical records (EMR), research data from the laboratories and enables parallelized computation on multiprocessor virtual machines. In POSEIDON, well-established computational environments, such as R, co-exist with custom-written pipelines. Daratumumab - an anti-CD38 therapeutic antibody - is a drug used for maintenance treatment of multiple myeloma. Daratumumab is administered after the stem cell transplant. Despite of its effectiveness, disease progression inevitably occurs after some time. This leads to patients not responding to daratumumab any longer. A Daratumumab sensitivity research trial is being conducted at COH, seeking to elucidate how multiple myeloma gains resistance to Daratumumab. Many patients undergo 20-22 treatment cycles, while some progress early and the cycles are stopped. Samples of peripheral blood of the patients on the trial are collected after each cycle of Daratumumab administration. All blood specimens are analyzed with CyTOF flow cytometry to determine and quantify cell populations. A fraction of samples is also analyzed with single-cell RNAseq to determine gene expression dynamics per population of cells. Finally, the laboratory values of each patient are added to the body of data. Given the multivariate longitudinal nature of the study as well as the sheer volume of the obtained data, a robust well-structured informatics system is needed. Here we present application of the POSEIDON environment to our ongoing trial. As shown on the Figure, POSEIDON is established within a cloud (AWS, Azure), which provides practically unlimited space for storage. The storage is redundantly distributed across multiple cloud nodes such that the data loss is unlikely. Also, any number of virtual machines may be deployed to perform computation. Specifically, the structured data are stored in POSEIDON in the SQLite database, which is in turn updated every two hours from the SQL Server located at COH. The SQL server at COH is populated by the laboratory with processed flow cytometry data and patient's records. R notebook(s) within POSEIDON communicate with the SQLite database to extract, transform and analyze the trial's data. The investigators from the collaborating laboratories (i.e., Hematology, Mathematical Oncology) are interacting with the notebooks to perform the analyses and present the results. In addition, the raw data are stored in POSEIDON as well. Therefore, POSEIDON establishes a single source of truth paradigm, which presents a single version of data to all involved parties. The cost associated with POSEIDON is twofold: computation and storage. Specifically, in our case a virtual machine of 4 CPU cores, 16 GB RAM and 128 GB solid-state storage space consumed $0.23 per hour when running. The storage cost for 14.6 GB of the project's data is $0.28 / month. In conclusion, it is important to note that POSEIDON can be scaled up beyond COH and beyond hematology.

IMPLEMENTASI RPG MAKER MV PADA GAME RPG TOWER OF ETERNITY

The development of technology has brought about changes in various aspects of human life, one of which is the entertainment media, namely video games. Video games are a common form of entertainment that typically follows certain rules, providing opportunities for achieving victory. Video games encompass various genres, one of which is RPG (Role-Playing Game), which usually emphasizes gameplay, story, and characters. The game "Tower of Eternity" was created using RPG Maker MV software version 1.6.2. The method used in the game's development is SDLC (System Development Life Cycle). The purpose of creating this game is for entertainment and to foster an interest in playing RPG genre games. The game has undergone testing in two stages: Functional testing and Device testing. Functional testing revealed that most of the buttons can be used effectively. Device testing on 5 devices, including a PC, Asus N551ZU laptop, Asus Vivobook X513EAN laptop, Asus VivoBook X421EAY laptop, and Toshiba Protege Z830 laptop, showed that 3 devices ran very well, 1 device ran reasonably well, and 1 device ran less efficiently. The required device specifications to run this game are AMD FX 7600P, 8GB RAM, and Radeon R9 M280x 4GB GDDR5.

A Methodology and Open-Source Tools to Implement Convolutional Neural Networks Quantized with TensorFlow Lite on FPGAs

Convolutional neural networks (CNNs) are used for classification, as they can extract complex features from input data. The training and inference of these networks typically require platforms with CPUs and GPUs. To execute the forward propagation of neural networks in low-power devices with limited resources, TensorFlow introduced TFLite. This library enables the inference process on microcontrollers by quantizing the network parameters and utilizing integer arithmetic. A limitation of TFLite is that it does not support CNNs to perform inference on FPGAs, a critical need for embedded applications that require parallelism. Here, we present a methodology and open-source tools for implementing CNNs quantized with TFLite on FPGAs. We developed a customizable accelerator for AXI-Lite-based systems on chips (SoCs), and we tested it on a Digilent Zybo-Z7 board featuring the XC7Z020 FPGA and an ARM processor at 667 MHz. Moreover, we evaluated this approach by employing CNNs trained to identify handwritten characters using the MNIST dataset and facial expressions with the JAFFE database. We validated the accelerator results with TFLite running on a laptop with an AMD 16-thread CPU running at 4.2 GHz and 16 GB RAM. The accelerator’s power consumption was 11× lower than the laptop while keeping a reasonable execution time.

Automatic Treatment Planning for Multi-focal Dynamic Conformal Arc GRID Therapy for Late-Stage Lung Cancer: A Feasibility Study

Palliative management of large, symptomatic pulmonary lesions, either as primary lung cancers or metastases, can be challenging due to need to balance effective radiation doses for cytoreduction with safety. Spatially Fractionated Radiation Therapy (SFRT), or GRID Therapy, is an emerging technique, which delivers ablative doses of radiotherapy to small, selected areas of tumor, while sparing organs-at-risk (OARs), and has been shown to debulk large lesions in preliminary studies. Conventionally, an alloy GRID block is manufactured to deliver GRID therapy. However, this delivery technique poses a challenge due to need for block, and dosimetrically when the tumor is deep-seated as excess dose may be delivered to OARs, such as skin and chest wall. This study aims to develop a fast, automatic planning solution using multi-focal dynamic conformal arcs (DCA) on modern Linear Accelerator. One late-stage lung cancer patient with simulated sphere target grid was included in this study. The sphere targets are 1.5cm in diameter and 4.3cm spacing. Four co-planar full arcs were used for optimization. The problem is formalized as finding optimal multi-leaf collimator (MLC) sequencing to cover N targets with K control points (CPs) for each arc. The state of each target's MLC opening at each CP is binary. In order to solve this NP-hard problem, the optimal solution was approximated by eliminating projection collision at each CP. MLC motion continuity and maximum speed were included in the cost function to ensure deliverability. The optimization started with randomized initial CP apertures, followed by solving state-transition equations for following CPs. Two grid arrays (9 and 10 targets respectively) were tested for plan quality. For each grid of target, the arc collimator angle was planned with 0 and 30 degrees for comparison. Prescription was 20 Gy per fraction. Monte Carlo simulation dose engine from matRad toolkit was used for dose calculation. Key dosimetric endpoints including target mean dose, D5%(Gy) and D95%(Gy), were reported. Average calculation time on the AMD Ryzen 5 5600 × 6-Core 3.7GHz CPU and 32GB RAM platform varied from 31 to 44 minutes. One zero-degree collimator and one thirty-degree collimator were generated for each target array. For nine-target array, mean target dose from both plans ranged from 23.41 to 26.55 Gy, while D5%(Gy) and D95%(Gy) ranged from 25.45 to 30.16 Gy, and 20.00 to 22.21 Gy, respectively. For ten-target array, the range of target mean, D5%(Gy) and D95%(Gy) were 23.82 to 28.74 Gy, 26.50 to 33.11 Gy, and 20.00 to 22.49 Gy. A fast, automatic planning solution for multi-focal DCA GRID therapy was developed. It provides clinically feasible plans with high efficiency for small target arrays for the late-stage cancer patient. The implementation provides excellent coverage for deep-seated tumors where alloy grid solution could fail to meet coverage objectives. Additional patients are needed in the future to further refine the technique.

Automated Georectification, Mosaicking and 3D Point Cloud Generation Using UAV-Based Hyperspectral Imagery Observed by Line Scanner Imaging Sensors

Hyperspectral sensors mounted on unmanned aerial vehicles (UAV) offer the prospect of high-resolution multi-temporal spectral analysis for a range of remote-sensing applications. However, although accurate onboard navigation sensors track the moment-to-moment pose of the UAV in flight, geometric distortions are introduced into the scanned data sets. Consequently, considerable time-consuming (user/manual) post-processing rectification effort is generally required to retrieve geometrically accurate mosaics of the hyperspectral data cubes. Moreover, due to the line-scan nature of many hyperspectral sensors and their intrinsic inability to exploit structure from motion (SfM), only 2D mosaics are generally created. To address this, we propose a fast, automated and computationally robust georectification and mosaicking technique that generates 3D hyperspectral point clouds. The technique first morphologically and geometrically examines (and, if possible, repairs) poorly constructed individual hyperspectral cubes before aligning these cubes into swaths. The luminance of each individual cube is estimated and normalised, prior to being integrated into a swath of images. The hyperspectral swaths are co-registered to a targeted element of a luminance-normalised orthomosaic obtained using a standard red–green–blue (RGB) camera and SfM. To avoid computationally intensive image processing operations such as 2D convolutions, key elements of the orthomosaic are identified using pixel masks, pixel index manipulation and nearest neighbour searches. Maximally stable extremal regions (MSER) and speeded-up robust feature (SURF) extraction are then combined with maximum likelihood sample consensus (MLESAC) feature matching to generate the best geometric transformation model for each swath. This geometrically transforms and merges individual pushbroom scanlines into a single spatially continuous hyperspectral mosaic; and this georectified 2D hyperspectral mosaic is then converted into a 3D hyperspectral point cloud by aligning the hyperspectral mosaic with the RGB point cloud used to create the orthomosaic obtained using SfM. A high spatial accuracy is demonstrated. Hyperspectral mosaics with a 5 cm spatial resolution were mosaicked with root mean square positional accuracies of 0.42 m. The technique was tested on five scenes comprising two types of landscape. The entire process, which is coded in MATLAB, takes around twenty minutes to process data sets covering around 30 Ha at a 5 cm resolution on a laptop with 32 GB RAM and an Intel® Core i7-8850H CPU running at 2.60 GHz.

Fuzzy-Based Active Queue Management Using Precise Fuzzy Modeling and Genetic Algorithm

Active Queue Management (AQM) methods significantly impact the network performance, as they manage the router queue and facilitate the traffic flow through the network. This paper presents a novel fuzzy-based AQM method developed with a computationally efficient precise fuzzy modeling optimized using the Genetic Algorithm. The proposed method focuses on the concept of symmetry as a means to achieve a more balanced and equitable distribution of the resources and avoid bandwidth wasting resulting from unnecessary packet dropping. The proposed method calculates the dropping probability of each packet using a precise fuzzy model that was created and tuned in advance and based on the previous dropping probability value and the queue length. The tuning process is implemented as an optimization problem formulated for the b0, b1, and b2 variables of the precise rules with an objective function that maximizes the performance results in terms of loss, dropping, and delay. To prove the efficiency of the developed method, the simulation was not limited to the common Bernoulli process simulation; instead, the Markov-modulated Bernoulli process was used to mimic the burstiness nature of the traffic. The simulation is conducted on a machine operated with 64-bit Windows 10 with an Intel Core i7 2.0 GHz processor and 16 GB of RAM. The simulation used Java programming language in Apache NetBeans Integrated Development Environment (IDE) 11.2. The results showed that the proposed method outperformed the existing methods in terms of computational complexity, packet loss, dropping, and delay. As such, in low congested networks, the proposed method maintained no packet loss and dropped 22% of the packets with an average delay of 7.57, compared to the best method, LRED, which dropped 21% of the packets with a delay of 10.74, and FCRED, which dropped 21% of the packets with a delay of 16.54. In highly congested networks, the proposed method also maintained no packet loss and dropped 48% of the packets, with an average delay of 16.23, compared to the best method LRED, which dropped 47% of the packets with a delay of 28.04, and FCRED, which dropped 46% of the packets with a delay of 40.23.