Accepted Papers

Investigation of Faster-RCNN Inception RESNET V2 on Off-line Kanji Handwriting Characters

Anthony A. Adole¹, Dr. Chris Bearchell² and Prof. Eran Edirisinghe¹, ¹Department of Computer Science, EPSRC Centre for Doctoral Training in Embedded Intelligence, Loughborough University, Leicestershire, UK, ²Surface Intelligence, Oxford, UK

ABSTRACT

In recent years detection and recognition of Off-line handwriting character has being a major task in the computer vision sector, researchers are looking on to developing deep learning models to avoid the traditional approaches which involves the tedious task of using the conventional methods for feature extraction and localization. However, state-of-the-art object detection modelsrely upon region proposal algorithms as a result, they settle for object locations principles, such network reduces thetime period of those detection network, exposing region proposal computation as a bottleneck. Faster-RCNN is a popular model used for recognition purpose in many recognition tasks, the goal of this paper is to serve as a guide for Multi-Classification on offline Handwriting Document using Pre-trained Faster-RCNN with inception resnet v2 feature Extractor. The result obtained from the experiments shows improved pre-trained models can be used insolving the research question concerning handwriting detection and recognition.

KEYWORDS

Offline Handwriting recognition and detection, faster-RCNN, inception resnet v2, Kanji handwriting, Japanese offline document, recognition and detection

An Inner/Outer Loop Ensemble-variational Data Assimilation Method

Yueqi Han^1,2, Bo Yang^1,2, Yun Zhang¹, Bojiang Yang¹ and Yapeng Fu^1,2, ¹College of Meteorology and Oceanography, National University of Defense Technology, Nanjing, China, ²National Key Laboratory on Electromagnetic Environmental Effects and Electro-optical Engineering, PLA Army Engineering University, Nanjing, China

ABSTRACT

Data assimilation (DA) for the non-differentiable parameterized moist physical processes is a complicated and difficult problem, which may result in the discontinuity of the cost function (CF) and the emergence of multiple extreme values. To solve the problem, this paper proposes an inner/outer loop ensemble-variational algorithm (I/OLEnVar) to DA. It uses several continuous sequences of local linear quadratic functions with single extreme values to approximate the actual nonlinear CF so as to have extreme point sequences of these functions converge to the global minimum of the nonlinear CF. This algorithm requires no adjoint model and no modification of the original nonlinear numerical model, so it is convenient and easy to design in assimilating the observational data during the non-differentiable process. Numerical experimental results of DA for the non-differentiable problem in moist physical processes indicate that the I/OLEnVar algorithm is feasible and effective. It can increase the assimilation accuracy and thus obtain satisfactory results. This algorithm lays the foundation for the application of I/OLEnVar method to the precipitation observational data assimilation in the numerical weather prediction (NWP) model.

KEYWORDS

Ensemble-variational Data Assimilation, Non-differentiable, Inner/Outer Loop

Using SDR Platform to Extract The RF Fingerprint of the Wireless Devices for Device Identification

Ting-Yu Lin, Chia-Min Lai and Chi-Wei Chen, Institute for Information Industry, Taipei, R.O.C

ABSTRACT

Due to the advent of the Internet of Things era, the number of related wireless devices is increasing, making the abundant and complex information networks formed by communication between devices. Therefore, security and trust between devices a huge challenge. In the traditional identification method, there are identifiers such as hash-based message authentication code, key, and so on, often used to mark a message that the receiving end can verify it. However, this kind of identifiers is easy to tamper. Therefore, recently researchers address the idea that using RF fingerprint, also called radio frequency fingerprint, for identification. Our paper demonstrates a method that extracts properties and identifies each device. We achieved a high identification rate, 99.9% accuracy in our experiments where the devices communicate with Wi-Fi protocol. The proposed method can be used as a stand-alone identification feature, or for two-factor authentication.

KEYWORDS

Internet-of-Things (IoT), Authentication, RF fingerprint, Machine Learning (ML), Device Identification

Change Detection using Synthetic Aperture Radar Videos

Hasara Maithree, Dilan Dinushka and Adeesha Wijayasiri, Department of Computer Science and Engineering, University of Moratuwa, Moratuwa, Sri Lanka

ABSTRACT

Many researches have been carried out for change detection using temporal SAR images. In this paper an algorithm for change detection using SAR videos has been proposed. There are various challenges related to SAR videos such as high level of speckle noise, rotation of SAR image frames of the video around a particular axis due to the circular movement of airborne vehicle, non-uniform back scattering of SAR pulses. Hence conventional change detection algorithms used for optical videos and SAR temporal images cannot be directly utilized for SAR videos. We propose an algorithm which is a combination of optical flow calculation using Lucas Kanade (LK) method and blob detection. The developed method follows a four steps approach: image filtering and enhancement, applying LK method, blob analysis and combining LK method with blob analysis. The performance of the developed approach was tested on SAR videos available on Sandia National Laborataries website and SAR videos generated by a SAR simulator.

KEYWORDS

Remote Sensing, SAR videos, Change Detection

Derivation of Loop Gain And Stability Test for Low-pass Tow-Thomas Biquad Filter

MinhTri Tran, Anna Kuwana and Haruo Kobayashi, Division of Electronics and Informatics, Gunma University, Kiryu 376-8515, Japan

ABSTRACT

Proposed derivation and measurement of self-loop function for a low-pass Tow Thomas biquadratic filter are introduced. The self-loop function of this filter is derived and analyzed based on the widened superposition principle. The alternating current conservation technique is proposed to measure the selfloop function. Research results show that the selected passive components (resistors, capacitors) of the frequency compensation of Miller’s capacitors in the operational amplifier and the Tow Thomas filter can cause a damped oscillation noise when the stable conditions for the transfer functions of these networks are not satisfied.

KEYWORDS

Superposition, Self-loop Function, Stability Test, Tow-Thomas Biquadratic Filter, Voltage Injection

Design of Active Inductor and Stability Test for Passive RLC Low-Pass Filter

MinhTri Tran, Anna Kuwana, and Haruo Kobayashi, Division of Electronics and Informatics, Gunma University, Kiryu 376-8515, Japan

ABSTRACT

Proposed stability test for RLC low-pass filters is presented. The self-loop functions of these filters are derived and analyzed based on the widened superposition principle. The alternating current conservation technique is proposed to measure the self-loop function. An active inductor is replaced with a general impedance converter. Our research results show that the values of the selected passive components (resistors, capacitors, and inductors) in these filters can cause a damped oscillation noise when the stable conditions for the transfer functions of these networks are not satisfied.

KEYWORDS

Widened Superposition, RLC Low-Pass Filter, Stability Test, Self-loop Function, Voltage Injection

A Spectrally Efficient MU-MIMO Turbo Receiver

Ye-Shun Shen, Fang-Biau Ueng* and Hung-Sheng Wang, Department of Electrical Engineering,National Chung-Hsing University, Taichung, Taiwan

ABSTRACT

Single carrier-frequency division multiple access (SC-FDMA) has been adopted as the uplink transmission standard in fourth generation cellular network to enable the power efficiency transmission in mobile station. Since multiuser multiple input multiple output (MU-MIMO) is a promising technology to fully exploit the channel capacity in mobile radio network, this paper investigates the uplink transmission of MU-MIMO SC-FDMA system with orthogonal space frequency block codes (SFBC). It is preferable to minimize the length of the cyclic prefix (CP) to improve the transmission energy and spectrum efficiency. Several techniques for block transmission without CP have been investigated. CP removal at the transmitter is compensated by a CP reconstruction at the receiver where only the past interference components are considered. In this paper, the chained turbo equalization with chained turbo estimation is employed in the designed receiver. The chained turbo estimation employs short training sequence (TR) that can improve the spectrum efficiency without sacrificing the estimation accuracy. In this paper, we propose a novel spectrally efficient iterative joint channel estimation, multiuser detection and turbo equalization for MU-MIMO SC-FDMA system without CP and with short TR. Some simulation examples for uplink scenario are given to demonstrate the effectiveness of the proposed scheme.

KEYWORDS

MU-MIMO SC-FDMA, chained turbo equalization, chained turbo estimation

Automated Classification of EEG Signals using Bagging and Boosting

Abdulhamit Subasi, Saeed Mian Qaisar, Effat University, College of Engineering, Jeddah, 21478, Saudi Arabia

ABSTRACT

In the cerebral surgery the positioning of epileptic foci is an elementary step. It is carried out by detecting the seizure in the electroencephalographic (EEG) recordings. In this framework, EEG Signals are composed of two classes, focal and non-focal. The focal signals are captured from brain areas in which the initial modifications to ictal EEG are sensed. The non-focal signals are recorded from brain areas that are not included at the seizure onset. A new focus area localization method is introduced based on various ensemble machine learning strategies and signal processing approach. The efficiency of the proposed method is assessed using classification accuracy, the area under the receiver operating characteristic (ROC) curve (AUC), and the F-measure. The system attains 98.8% accuracy. It confirms the potential of using the proposed solution in modern EEG analysis systems.

KEYWORDS

Electroencephalogram (EEG), Auto regressive (AR) method, Ensemble Machine Learning Methods.

Tala Classification in Carnatic Music using Audio Thumbnailing

Amulya Sri Pulijala and Suryakanth V Gangashetty, International Institute of Information Technology, Hyderabad, India

ABSTRACT

The concept of Raga and Tala is integral part of Indian Classical music. Raga is the melodic component while Tala is the rhythmic component in the music. Hence, Tala classification and identification is a paramount problem in the area of Music Information Retrieval (MIR) systems. Although there are seven basic Talas in Carnatic Music, a further subdivision of them gives a total of 175 ragas. Statistical and machine learning approaches are proposed in Literature Survey to classify Talas. However, they use complete musical recording for training and testing. As part of this paper, a novel approach is proposed for the first time in Carnatic music to classify Talas using repetitive structure called Thumbnails.

KEYWORDS

Tala Classification, Carnatic Music, Audio Thumbnails, SVM, CNN.

Deep Learning based Classification of 2D and 3D Images for Facial Expression Recognition: Comparison Study

Fouzia Adjailia, DianaOlejarova and Peter Sincak, Dep. Cybernetics and Artificial Intelligence, Technical University of Kosice, Kosice, Slovak Republic

ABSTRACT

Facial expressions are an important communication channel among human beings. The Classification of facial expressions is a research area which has been proposed in several fields in recent years, it provides insight into how human can express their emotions which can be used to inform and identify a person's emotional state. In this paper, we provide the basic outlines of both two dimensional and three-dimensional facial expression classification with a number of concepts in detail and the extent of their influence on the classification process. We also compare the accuracy of two-dimensional (2D) and three-dimensional (3D) proposed models to analyse the 2D and 3D classification using comprehensive algorithms based on convolution neural network, the model was trained using a commonly used dataset named Bosphorus. Using the same experimental setup, we discussed the results obtained in terms of accuracy and set a new challenge in the classification of facial expression.

KEYWORDS

Convolution neural network, facial expression classification, bosphorus, voxel classification.

Fast Prototyping Executable Model for Distributed Embedded Systems

Tubonimi Jenewari and David Mulvaney, Mechanical, Electrical and Manufacturing Engineering, Loughborough University, Loughborough, UK

ABSTRACT

The aim of the research is to provide a framework for prototyping executable model for distributed embedded system, it includes both hardware and software, in this way the modelling and implementation process will allow seamless execution of distributed system at functional, hardware simulation and hardware realization levels. An example of the setup will be an ABS (anti-lock braking system) that comprises of system components; sensor, actuator, ECU’s which are interconnected through the vehicle communication bus. Software automobile vehicle model will provide the vehicle dynamics and the model of the ABS that will mimic the functionality of ABS. Multiple processors (ECU) will be interconnected in a distributed format to reduce the time of execution.

All this is captured in a Unified Modelling Language (UML) a standard for the idea of the design to be captured, expressive enough to be understood. The UML notation will be converted to Extensible Markup Language (XML), and a parser code written to extract all the necessary class information (to get the classes where the simulation code will come in) from the overwhelming lines of XML to be transferred to the ECU for execution of the simulation. The simulation is carried out using GDB.

KEYWORDS

Prototyping, distributed embedded system, executable model, UML

Ready to Use Virtual Machine Pool Cache usingWarm Cache

Sudeep Kumar, Deepak Kumar Vasthimal, and Musen Wen, eBay Inc., 2025 Hamilton Ave, San Jose, CA 95125, USA

ABSTRACT

Today, a plethora of distributed applications are managed on internally hosted cloud platforms. Such managed platforms are often multi tenant by nature and not specifically tied to a single use-case. Smaller footprint of infrastructure on a managed cloud platform has its own set of challenges especially when applications are required to be infrastructure aware for quicker deployments and response times. There are often times and challenges to quickly spawn ready to use instances or hosts on such infrastructure. As part of this paper we outline mechanisms to quickly spawn ready to use instances for application while also being infrastructure aware. In addition, paper proposes architecture that provides high availability to deployed distributed applications.

KEYWORDS

cloud computing, virtual machine, elastic, elastic search, consul, cache, java, kibana, mongoDB, high performance computing, architecture.

Optimization of Multimodal Transport Organization based on Logistics Cloud Platform

Liqun Ding, School of Transportation, Wuhan University of Technology, Wuhan, Hubei, 430063 & Logistics College, Wuhan Technical College of Communications, Wuhan, Hubei, 430065, Chennai

ABSTRACT

Aiming at the problem of multi-modal Transportation Organization optimization, making full use of the advantages of cloud platform artificial intelligence technology and big data, the general idea, process and principle of multi-modal transportation dispatch strategy are designed,The primary consideration in the top-level design of multimodal transport system is the matching degree and coordination degree of port facilities and transport information platform. In the process of container transportation, there exists a mixed time window consisting of hard time window and soft time window. The scale effect of transportation is the result of a combination of internal and external factors. This section will study the coordination of transport organization considering scale effect under the constraint of mixed time window. This paper will elaborate the functions and operation status of the relatively independent information system of railway and port, and try to establish an electronic platform suitable for the information interconnection and interoperability of multimodal transport stations in combination with the traditional information exchange mode.

KEYWORDS

Logistics, Cloud Platform, information system, collaborative, operation.

Synchronization Aspects in 5G

Mridula Korde, Department of Electronics and Communication Engineering, Shri Ramdeobaba College of Engineering and Management, Nagpur, India

ABSTRACT

Increasing Internet data traffic has driven the capacity demands for currently deployed 3G and 4G wireless technologies. Now, intensive research toward 5th generation wireless communication networks is progressing in many fronts. 5G technologies are expected to be in use around 2020. Moving toward 5G, network synchronization is expected to play a key role in the successful deployment of the new mobile communication networks. Synchronization is an essential prerequisite for all mobile networks to operate. It’s fundamental to data integrity, and without it data will suffer errors and networks can suffer outages. ‘Loss of synchronization’ problems can be difficult to diagnose and resolve quickly and add to operational costs. Poor synchronization affects customer satisfaction and is therefore revenue affecting too. This paper presents synchronization requirement and related aspects in upcoming 5G technologies.

KEYWORDS

5G, MIMO, Synchronization