next day
all days

View: session overviewtalk overview

07:00-09:00 Session 1: FRINGE TALKS
FRINGE: Torus heat equations to get dizzy from. Hooping with xC
PRESENTER: Øyvind Teig

ABSTRACT. The story on how points representing temperature on a torus, in the form of occam PROCs running on one or several transputers, exchanging temperature with their neighbours, caught the interest of whether - or how this problem may be run as xC tasks on the XMOS xCORE multicore architecture.

Pizza Model - FPGA SME Transputer platform and the Path Forward
PRESENTER: Lindsay Quarrie

ABSTRACT. “What Kind of Transputers would you like on top of your FPGA” Virtual Pizza Break Out Session

Exchange of thoughts on what could go on next gen FPGA transputer development

and test platforms

09:00-12:00 Session 2: Power, Image Recognition and Language Processing Applications
Harmonic Elimination based on selective optimization for Capacitor Voltages Balancing in Multilevel Inverters with Considering Load power factor

ABSTRACT. In this paper, SHE based on Newton Raphson algorithm is investigated for flying capacitor multilevel inverter control. Without measuring current and capacitor voltages, as well as not using any controller and voltage feedback the proposed strategy is capable of determining the best switching angles pattern in low switching frequency for maintaining nil mean current capacitor DC-link, hence balancing the capacitor voltages and eliminating certain harmonic orders while considering inverter R,L load power factor variation. Theoretical results are verified by simulations and experiments. Results show that the proposed method does effectively eliminate specific low order harmonics and the output voltage results in low total harmonic distortion with balanced flying capacitors.

Real-Time Localization of Vehicle License Plate using Improved Faster Region-Based Convolutional Neural Network

ABSTRACT. Real-time identification of vehicle license plate has become more practical within the last decade in many applications such as; storage and retrieval of vehicular inflow records, automatic toll collection, parking fee payment, traffic monitoring, tracking of moving vehicles, recovery of stolen vehicles etc. A lot of researches have been carried out on vehicles license plates detection and localization which has led to the development of new techniques and modification of existing techniques. In this paper, we present an improved Faster Region-based Convolutional Neural Network (R-CNN) for localizing vehicle license plates. The Faster R-CNN utilized inceptionV2 architecture. The region of interest pooling of the existing Faster R-CNN was replaced by region of interest align which improved the creation quality of region proposals for license plates localization. The performance of the model was evaluated using Mean Average Precision (mAP) obtained from the precision-recall (PR) curves that were computed during model training and accuracy of 99% was achieved. The performance of the system was evaluated via real-time testing with 100 vehicles and localization accuracy of 99% was achieved. An effective camera to vehicle distance was also established via real time testing at different camera to vehicle distances.

Transfer learning using Pre-trained AlexNet for Marathi Handwritten Compound Character Image Classification

ABSTRACT. Transfer learning use to train the data faster and avoid over-fitting when the size of the dataset is small. The purpose of this work is to investigate Handwritten Marathi Compound Characters and Handwritten Marathi 0-9 digits using Pre-trained Convolutional Neural Network. In this article, AlexNet mainly used to train the handwritten compound characters of Marathi Script for image classification. The tests were conducted on Marathi Handwritten characters with 3,283 sample images of three compound characters, 400 images of two compound character and 1,000 Marathi Handwritten numbers with some preprocessing of resizing images into 227x 227 pixels. For three combining characters AlexNet gives the maximum accuracy of 96.77%, 97.3% and for Marathi handwritten digits gives 100% accuracy.