Hermes José Loschi Proceedings of the 3rd Brazilian Technology … · 2018. 10. 16. · Yuzo Iano · Rangel Arthur Osamu Saotome · Vania Vieira Estrela Hermes ... Lucas Heitzmann

Yuzo Iano · Rangel Arthur  Osamu Saotome · Vania Vieira Estrela  Hermes José Loschi   Editors

Proceedings of the 3rd Brazilian Technology SymposiumEmerging Trends and Challenges in Technology

Proceedings of the 3rd Brazilian TechnologySymposium

Yuzo Iano • Rangel ArthurOsamu Saotome • Vania Vieira EstrelaHermes José LoschiEditors

Proceedings of the 3rdBrazilian TechnologySymposiumEmerging Trends and Challengesin Technology

123

EditorsYuzo IanoFaculty of Electricaland Computer Engineering

University of CampinasCampinas, São PauloBrazil

Rangel ArthurFaculty of TechnologyUniversity of CampinasCampinas, São PauloBrazil

Osamu SaotomeDivisão de Engenharia EletrônicaInstituto Tecnológico de Aeronáutica (ITA)São José dos Campos, São PauloBrazil

Vania Vieira EstrelaUniversidade Federal Fluminense (UFF)Duque de Caxias, Rio de JaneiroBrazil

Hermes José LoschiFaculty of Electrical and ComputerEngineering

University of CampinasCampinas, São PauloBrazil

ISBN 978-3-319-93111-1 ISBN 978-3-319-93112-8 (eBook)https://doi.org/10.1007/978-3-319-93112-8

Library of Congress Control Number: 2018943254

© Springer International Publishing AG, part of Springer Nature 2019This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or partof the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations,recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmissionor information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilarmethodology now known or hereafter developed.The use of general descriptive names, registered names, trademarks, service marks, etc. in thispublication does not imply, even in the absence of a specific statement, that such names are exempt fromthe relevant protective laws and regulations and therefore free for general use.The publisher, the authors and the editors are safe to assume that the advice and information in thisbook are believed to be true and accurate at the date of publication. Neither the publisher nor theauthors or the editors give a warranty, express or implied, with respect to the material contained herein orfor any errors or omissions that may have been made. The publisher remains neutral with regard tojurisdictional claims in published maps and institutional affiliations.

This Springer imprint is published by the registered company Springer Nature Switzerland AGThe registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland

Foreword

It is with deep satisfaction that I write this Foreword to the Proceedings of the 3rdBrazilian Technology Symposium: Emerging Trends and Challenges inTechnology (BTSym’17), held at the University of Campinas-SP, Brazil, inDecember 2017. This event is in its third edition and has consolidated to become anexcellent opportunity for researchers, professors and students to present and discussthe results of their research works.

The 2017 edition of BTSym is characterised by the broad scope of the areasexposed, with papers dealing with current and essential topics for Brazilian andworld technological development, including subjects related to the various branchesof engineering, architecture and computer science.

Events such as BTSym are an essential part of the research and innovationprocess. First, these events contribute to the promotion of research activities, whichare key to a country’s technological development. The dissemination of researchresults, as promoted by BTSym, contributes to the transformation of researchfindings into technological innovation. In addition, these events facilitate thesharing of findings, leading eventually to the formation of research networks, whichaccelerate the achievement of new results. Therefore, I would like to congratulatethe BTSym General Chair, Prof. Yuzo Iano and his group of collaborators for theimportant initiative of organising the BTSym 2017 and for providing the oppor-tunity for authors to present their work to a wide audience through this publication.Last but not least, I congratulate the authors for the quality of the work presented inthis Proceedings.

Campinas, Brazil Prof. Paulo CardieriFormer President of the Brazilian

Telecommunications Society (2014–2017)University of Campinas

v

Preface

This book contains the refereed Proceedings of the 3rd Brazilian TechnologySymposium: Emerging Trends and Challenges in Technology, held in Campinas-SP,Brazil, in December 2017.

The Brazilian Technology Symposium is an excellent forum for presentationsand discussions of the latest results of projects and development research, in severalareas of knowledge, in scientific and technological scope, including Smart Designs,Sustainability, Inclusion, Future Technologies, Architecture and Urbanism,Computer Science, Information Science, Industrial Design, Aerospace Engineering,Agricultural Engineering, Biomedical Engineering, Civil Engineering, Control andAutomation Engineering, Production Engineering, Electrical Engineering,Mechanical Engineering, Naval and Oceanic Engineering, Nuclear Engineering,Chemical Engineering, Probability and Statistics.

This event seeks to bring together researchers, students and professionals fromthe industrial and academic sectors, seeking to create and/or strengthen the linkagesbetween issues of joint interest. Participants were invited to submit research paperswith methodologies and results achieved in scientific level research projects,completion of coursework for graduation, dissertations and theses.

The 32 full papers accepted for this book were selected from 102 submissions, andin each case, the authors were shepherded by an experienced researcher, with arigorous peer-review process. Among the main topics covered in this book, we canhighlight Artificial Neural Networks, Computational Vision, Security Applications,Web Tool, Cloud Environment, Network Functions Virtualization, Software-DefinedNetworks, IoT, Residential Automation, Data Acquisition, Industry 4.0, Cyber-Physical Systems, Digital Image Processing, Infrared Images, Patters Recognition,Digital Video Processing, Precoding, Embedded Systems, Machine Learning,Remote Sensing, Wireless Sensor Network, Heterogeneous Networks, UnmannedGround Vehicle, Unmanned Aerial System, Security, Surveillance, Traffic Analysis,Digital Television, 5G, Image Filter, Partial Differential Equation, Smoothing Filters,Voltage Controlled Ring Oscillator, Difference Amplifier, Photocatalysis,Photodegradation, Cosmic Radiation Effects, Radiation Hardening Techniques,Surface Electromyography, Sickle Cell Disease Methodology, MicroRNAs, Image

vii

Processing Venipuncture, Cognitive Ergonomics, Ecosystem Services,Environmental, Power Generation, Ecosystem Services Valuation, Solid Waste andUniversity Extension.

We hope you enjoy and take advantage of this book and feel motivated to submitus your papers in the future to Brazilian Technology Symposium.

Best wishes,

Campinas, Brazil Prof. Hermes José LoschiTechnical and Finance Chair of Brazilian

Technology Symposium

viii Preface

Organizing Committee

Organizing and Executive Committee

Yuzo Iano, LCV/DECOM/FEEC/UNICAMP, General Chair BTSym & WSGEOsamu Saotome, ITA, Associate-General Chair BTSymRangel Arthur, FT/UNICAMP, Vice-General Chair BTSymEvaldo Gonçalves Pelaes, UFPA, Vice-Associate-General Chair BTSymHermes José Loschi, LCV/DECOM/FEEC/UNICAMP, Technical Program andFinance ChairTelmo Cardoso Lustosa, LCV/DECOM/FEEC/UNICAMP, Local ArrangementsChairCamila Santana Domingues, LCV/DECOM/FEEC/UNICAMP, Registration ChairLuiz Antonio de Sousa Ferreira, LCV/DECOM/FEEC/UNICAMP, ProceedingsChairReinaldo Padilha, LCV/DECOM/FEEC/UNICAMP, Communication ChairAna Carolina Borges Monteiro, LCV/DECOM/FEEC/UNICAMP, Marketing ChairDouglas Aguiar do Nascimento, LCV/DECOM/FEEC/UNICAMP, InstitucionalRelationship Chair

Scientific and Academic Committee

Osamu Saotome, ITAVania Vieira Estrela, UFF, Rio de JaneiroLuiz Cézar Martini, DECOM/FEEC/UNICAMPDavid Bianchini, PUC/CampinasLuis Geraldo Pedroso Meloni, DECOM/FEEC/UNICAMPAna Cláudia Seixas, PUC/Campinas

ix

Cristiano Akamine, Universidade Presbiteriana MackenzieLuiz Vicente Figueira de Mello Filho, Universidade Presbiteriana MackenzieGuillermo Leopoldo Kemper Vásquez, USMP & UNI-INICTELLucas Heitzmann Gabrielli, DECOM/FEEC/UNICAMPEdgard Luciano Oliveira da Silva, EST/UEATalía Simões dos Santos, FT/UNICAMPJanito Vaqueiro Ferreira, DMC/FEM/UNICAMPVlademir de Jesus Silva Oliveira, UNEMAT/SinopHugo Enrique Hernandez Figueroa, DECOM/FEEC/UNICAMPMarcos Antonio do Nascimento Guimarães, UNIP/CAMPINAS, JUNDIAÍMaria Thereza de Moraes Gomes Rosa, Universidade Presbiteriana MackenzieAngela del Pilar Flores Granados, FEA/UNICAMPPaolo Bacega, Faculdade AnhangueraMarcos Fernando Espindola, IFSP São PauloPolyane Alves Santos, Instituto Federal Da BahiaJude Hemanth, Department of Electrical and Computer Engineering, KarunyaUniversity, Coimbatore, India

Technical Reviewers Committee

Adão Boava, Universidade Federal de Santa Catarina, UFSCAna Carolina Borges Monteiro, LCV/DECOM/FEEC/UNICAMPAmilton Lamas, PUC-CampinasAgord de Matos Pinto Júnior, LCV/DECOM/FEEC/UNICAMPAngela del Pilar Flores Granados, FEA/UNICAMPSilvio Renato Messias de Carvalho, LCV/DECOM/FEEC/UNICAMPJoaquim Marcos Santa Rita da Silva, Instituto Nacional de TelecomunicaçõesJosé Alexandre Nalon, Centro Universitário Salesiano São Paulo, UNISALMurilo Cesar Perin Briganti, LCV/DECOM/FEEC/UNICAMPLuigi Ciambarella Filho, Universidade Veiga de Almeida/Develop BiotechnologyIngrid Araujo Sampaio, Universidade Estadual de CampinasHermes José Loschi, LCV/DECOM/FEEC/UNICAMPDaniel Rodrigues Ferraz Izario, LCV/DECOM/FEEC/UNICAMPMariana Carvalho, DCA/FEEC/UNICAMPDiego Pajuelo, LCV/DECOM/FEEC/UNICAMPDouglas Aguiar do Nascimento, FACTI/LCV/DECOM/FEEC/UNICAMPEdson José Gonçalves, LCV/DECOM/FEEC/UNICAMPMarcos Fernando Espindola, IFSP São PauloPolyane Alves Santos, Instituto Federal Da BahiaRangel Arthur, INOVA/FT/UNICAMPReinaldo Padilha, LCV/DECOM/FEEC/UNICAMP

x Organizing Committee

Kelem Christine Pereira Jordão, LCV/DECOM/FEEC/UNICAMPEuclides Lourenço Chuma, LCV/DECOM/FEEC/UNICAMPJosé Yauri, DCA/FEEC/UNICAMPJulío Humberto León Ruiz, LCV/DECOM/FEEC/UNICAMP

Sponsor

Beta Telecommunications

Organizing Committee xi

Acknowledgements

Our appreciation goes to a lot of colleagues and friends who assisted in thedevelopment of this book: Proceedings of the 3rd Brazilian TechnologySymposium: Emerging Trends and Challenges in Technology.

First of all, I would like to thank all the members of the Organizing and ExecutiveCommittee, for the commitment throughout the year, several meetings were held andmany challenges were overcome for the accomplishment of the BTSym 2017. Also,and with great merit, I would like to thank all the members of the Scientific andAcademic Committee, and Technical Reviewers Committee for their excellent work,which was essential to ensure the quality of our peer-review process, and collabo-rating with the visibility and technical quality of the BTSym 2017.

The Brazilian Technology Symposium is an event created by Laboratory ofVisual Communications of the Faculty of Electrical and Computer Engineeringof the University of Campinas (UNICAMP). In this way, I would like to thank theUNICAMP, especially the UNICAMP Cultural Development Center for the supportand hosting of the BTSym 2017, which was fundamental for the successful theiraccomplishment.

Beta Telecommunications played a key role in holding the BTSym 2017; due tothe financial support from it, it was possible to consolidate with quality manyBTSym 2017 organization aspects, which ensured the quality to support the authorsand Speakers.

Finally, I thank all the authors for their participation in the BTSym 2017, Isincerely hope to have provided an experience that was very useful and enriching inthe personal and professional life of everyone, and my special thanks go to Profa.Vania Vieira Estrela. In my almost 50 years of academic career in UNICAMP, few

xiii

were the opportunity to work with a professional such her. She is definitely one of akind, extremely efficient, hardworking and the BTSym 2017 certainly has much tothank for Prof. Vania Vieira Estrela.

Best wishes,

Prof. Yuzo IanoGeneral Chair of Brazilian Technology Symposium

xiv Acknowledgements

Contents

Cloud Detection for PERUSAT-1 Imagery Using Spectral and TextureDescriptors, ANN, and Panchromatic Fusion . . . . . . . . . . . . . . . . . . . . . 1Giorgio Morales, Samuel G. Huamán and Joel Telles

Implementation of a Digital Image Processing Algorithm on a ColibriIMX6 Embedded Industrial System for Optical Mark Recognition . . . . 9Carlos Herrera, Stephany Del Campo, Abel Dueñas, Julio León,Guillermo Kemper and Christian del Carpio

IEEE 802.11 De-authentication Attack Detection Using MachineLearning on Unmanned Aerial System . . . . . . . . . . . . . . . . . . . . . . . . . . 23Gustavo de Carvalho Bertoli and Osamu Saotome

Detection and Visualization of Forearm Veins for Venipuncture Basedon Digital Processing of Infrared Image . . . . . . . . . . . . . . . . . . . . . . . . . 31Kevin Espinoza, Bryan Magallanes and Guillermo Kemper

Measuring the Level of Mildew in Quinoa Plantations Based on DigitalImage Processing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43Gian Oré, Alexis Vásquez, Guillermo Kemper and Jonell Soto

An Architecture for Flow-Based Traffic Analysis in a CloudEnvironment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55Tiago Primini, Eric Baum, Leonardo Mariote, Matheus Ribeiroand Giovanni Curiel

Development of Residential Automation Modules for Performing LowComplexity Activities Using IoT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67Lahis G. de Almeida, Rachel B. de Lima and Edgard Luciano O. da Silva

Internet of Things: An Overview of Architecture, Models,Technologies, Protocols and Applications . . . . . . . . . . . . . . . . . . . . . . . . 75J. R. Emiliano Leite, Paulo S. Martins and Edson L. Ursini

xv

Evaluation of Traffic Delays and Utilization of IoT NetworksConsidering Mobility . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87J. R. Emiliano Leite, Edson L. Ursini and Paulo S. Martins

Blocking of the Cell Overflow Traffic in Heterogeneous Networks . . . . . 95Loreno M. Silveira, Paulo S. Martins and Edson L. Ursini

Digital Image Processing with Data Storagefor Security Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103Daniel Izario, Yuzo Iano, Bruno Izario and Diego Castro

Development of a Digital Image Processing Web Tool for a MonitoringSystem Relying on an Unmanned Ground Vehicle . . . . . . . . . . . . . . . . . 111Daniel Izario, Yuzo Iano, Bruno Izario, Letícia Magalhãesand Diego Castro

Edge-Detection Noise-Smoothing Image Filter Techniques . . . . . . . . . . . 117Daniel Izario, Yuzo Iano, Bruno Izario, Diego Castro and Carlos Nazareth

Conception of an Electric Vehicle’s Robotic Platform Developedfor Applications on CTS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123Willian Gomes de Almeida, Juliano de Almeida Monte-Mor,Rafael Francisco dos Santos, Eben-Ezer Prates da Silveira,Sandro Carvalho Izidoro, Tarcísio Gonçalves de Brito,Natália Cosse Batista and Giovani Bernardes Vitor

Pulse Shaping Filter Design for Filtered OFDM Transceivers . . . . . . . . 131Jaime J. Luque Quispe and Luís G. Pedroso Meloni

Digital TV (ISDB-Tb) Broadcasting over LTE Broadcast:A Feasibility Study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145Cibele A. Makluf, Julio León and Yuzo Iano

A Packet Scheduling Algorithm with Traffic Policingin LTE Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157Jeanette Quiñones Ccorimanya and Lee Luan Ling

Programmable Data Plane with Stateful Flow Processingfor NFV Acceleration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169Marcelo Nascimento, Tiago Primini, Eric Baum, Pedro Martucci,Francisco Cabelo and Leonardo Mariote

Design of the Voltage-Controlled Ring Oscillator Using OptimizationTools (MunEDA® WiCkeD) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179Agord de Matos Pinto Jr., Raphael R. N. Souza, Leandro Tiago Manera,Jorge Enrique Vargas Solano, Cássia Maria Chagas and Saulo Finco

xvi Contents

Computational Performance of an Model for WirelessTelecommunication Systems with Discrete Eventsand Multipath Rayleigh . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 193Reinaldo Padilha, Yuzo Iano, Edson Moschim,Ana Carolina Borges Monteiro and Hermes José Loschi

Electrical Power Monitoring of Low-Power DevicesUsing a Smart Meter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205Rachel Batalha de Lima and Edgard Luciano Oliveira da Silva

Applied Instrumentation: Strain Measurements Using Arduinoand Strain Gauge . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 213Elton Fernandes dos Santos, Vlademir de Jesus Silva Oliveira, Wagnerde Almeida Ferreira and Julio César Beltrame Benatti

Overview About Radiation–Matter Interaction Mechanisms andMitigation Techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 223R. N. S. Raphael, L. E. Seixas Jr., Agord M. Pinto Jr., S. A. Bascopé,L. T. Manera, S. Finco and S. P. Gimenez

Biopotential Amplification System Developed for SurfaceElectromyography Using Dry Electrodes . . . . . . . . . . . . . . . . . . . . . . . . 239Alex Toshio Kakizaki, Marcos Henrique Mamoru Otsuka Hamanaka,Vinícius do Lago Pimentel, Carlos Alexandre Ferriand Antônio Augusto Fasolo Quevedo

E-Street for Prevention of Falls of the Elderly an Urban VirtualEnvironment for Human–Computer Interaction from Lower LimbMovements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 249Alexandre Brandão, Diego Dias, Iago Alvarenga, Glesio Paiva,Luis Trevelin, Karina Gramany-Say and Gabriela Castellano

A Time Series Analysis Applied to the Generation of Energy at theSanto Antonio Hydroelectric Plant Located in the State of Rondoniain Brazil in the Year 2016 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 257Nickolas Bastiani Cassiano, Joao Gabriel Ribeiro, Giovane Maia do Valeand Vlademir de Jesus Silva Oliveira

Optimization of Photocatalytic Degradation of Methyl OrangeUsing TiO2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 269Gustavo Duran Iga, André Luis de Castro Peixotoand Ademir Geraldo Cavallari Costalonga

Cognitive Ergonomics and the Industry 4.0 . . . . . . . . . . . . . . . . . . . . . . 275Alessandra Cristina Santos Akkari, Mateus Faraj Marques da Rochaand Rosani Franco de Farias Novaes

Contents xvii

The Contribution of the University Extension for Solid WasteManagement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 281Ana Claudia Mendes de Seixas and Giovanna Ramos Maccari

Using i-Tree Canopy to Estimate and Value Ecosystem Servicesof Air Pollutant Removal . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 291Tatiane Ferreira Olivatto

Detecting and Counting of Blood Cells Using Watershed Transform:An Improved Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 301Ana Carolina Borges Monteiro, Yuzo Iano and Reinaldo Padilha França

General Aspects of Pathophysiology, Diagnosis, and Treatmentof Sickle Cell Disease . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 311Ana Carolina Borges Monteiro, Yuzo Iano and Reinaldo Padilha França

Emergency Response Cyber-Physical System for Flood Preventionwith Sustainable Electronics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 319Vania V. Estrela, Jude Hemanth, Osamu Saotome, Edwiges G. H. Grataand Daniel R. F. Izario

xviii Contents

Cloud Detection for PERUSAT-1 ImageryUsing Spectral and Texture Descriptors,ANN, and Panchromatic Fusion

Giorgio Morales , Samuel G. Huamán and Joel Telles

Abstract The cloud detection process is a prerequisite for many remote sensingapplications in order to use only those cloud-free parts of satellite images and reduceerrors of further automatic detection algorithms. In this paper, we present amethod todetect clouds in high-resolution images of 2.8mper pixel approximately. The processis performed over those pixels that exceed a defined threshold of blue normalizeddifference vegetation index to reduce the execution time. From each pixel, a set oftexture descriptors and reflectance descriptors are processed in an Artificial NeuralNetwork. The texture descriptors are extracted using the Gray-Level Co-occurrenceMatrix. Each detection result passes through a false-positive discard procedure onthe blue component of the panchromatic fusion based on image processing tech-niques such as Region growing, Hough transform, among others. The results showa minimum Kappa coefficient of 0.80 and an average of 0.94 over a set of 25 imagesfrom the Peruvian satellite PERUSAT-1, operational since December 2016.

Keywords Cloud detection · High-resolution · Artificial neural networksTexture analysis

1 Introduction

In the context of the recent launch of the Peruvian satellite PERUSAT-1, several appli-cations in the areas of agriculture, environmental monitoring, cartography and secu-rity have been proposed, taking advantage of the high-resolution that these images

G. Morales (B) · S. G. Huamán · J. TellesNational Institute of Research and Training at Telecommunications (INICTEL-UNI),National University of Engineering, San Luis 1771, Lima 15021, Perue-mail: [email protected]

S. G. Huamáne-mail: [email protected]

J. Tellese-mail: [email protected]

© Springer International Publishing AG, part of Springer Nature 2019Y. Iano et al. (eds.), Proceedings of the 3rd Brazilian Technology Symposium,https://doi.org/10.1007/978-3-319-93112-8_1

1

2 G. Morales et al.

can provide. However, the presence of clouds in acquired satellite images can makethese analyses difficult. This is why, after atmospheric correction, cloud detectionis an important preliminary step for any subsequent study. Previous works on clouddetection involve multi-temporal analysis of images with close acquisition dates [1],threshold-based cloud detection [2, 3], extraction of spectral and texture informationusing intelligent classifiers [4, 5] and even more recent methods that propose the useof convolutional neural networks [6].

In this paper, we propose a methodology for the detection of clouds in high-resolutionmultispectral images of PERUSAT-1 using spectral and texture descriptorsprocessed by an Artificial Neural Network and discarding false positives by morpho-logical analysis on a panchromatic fusion, as shown in Fig. 1. Spectral informationis extracted from the red, green, blue, and near-infrared (NIR) bands; besides, wehighlight the use of the blue normalized difference vegetation index (BNDVI) [7].Spectral information is often not sufficient for classification tasks, so we will takeadvantage of the spatial information provided by high-resolution images [8]. The tex-ture descriptors are extracted using the Gray-Level Co-occurrence Matrix. The restof the paper is organized as follows. Section 2 explains the proposed cloud detectionmethodology; Sect. 3, the results obtained and Sect. 4 summarizes the conclusionsof the work.

2 Proposed Method

2.1 Experimental Data

A PERUSAT-1 image has four spectral bands: red (0.63–0.7 µm), green(0.53–0.59 µm), blue (0.45–0.50 µm) and NIR (0.752–0.885 µm). The spatial res-olution of the multispectral bands is 2.8 m per pixel and the panchromatic band,0.7 m per pixel. A total number of 32 composite images of variable area and fromdifferent geographies (e.g., rainforest, desert, agriculture and urban areas) have beenused: seven for the training and validation of the ANN and 25 for the validation ofthe method.

Fig. 1 Cloud detection procedure

Cloud Detection for PERUSAT-1 Imagery Using Spectral … 3

2.2 Feature Extraction

The descriptor vector X is composed of the following spectral and texture features:

Spectral and color descriptors. Since there are four spectral bands, the four DNvalues of each pixel are part of the cloud descriptor. In addition, three more featuresare added to the descriptor: H (Hue), S (Saturation) and V (Value). Furthermore,we consider BNDVI [7] as an additional descriptor taking into account that cloudshave relatively high and constant reflectance values between the blue and NIR bands,whichmeans that their BNDVI index would be lower than those of other objects suchas vegetation or silt loam.Texture descriptors. We calculate the texture descriptors from the Gray-Level Co-occurrence Matrix (GLCM), which is calculated from the blue band, where there ismore contrast between clouds and soil. Six statistical indicators are extracted fromthe GLCM matrix: Mean, variance, contrast, correlation, energy, and homogeneity[6] using a kernel size of 10 pixels for four directions: 0°, 45°, 90° and 135°, whichsums a total of 24 texture descriptors.

2.3 Artificial Neural Network Training

First, we create a data-set which consists of 22,386,038 descriptor vectors as patterns(‘clouds’ and ‘non-clouds’) extracted from seven manually annotated images of the32 available images. Then, we split 90% of the data to create the training set, 5% tothe validation set and 5% to the test set. To select an optimal model, we perform avariable selection step and a configuration selection step:

Feature selection. Joining the spectral and texture features, the X descriptor has adimension of 32 elements. We chose a neural network with two hidden layers ofseven and two neurons, respectively, to initialize a training model with all the 32variables available. Then, we remove one variable at a time and perform 10 trainingsto evaluate the causality. Finally, it is verified that when removing the ‘correlation’variable the changes in the identification precision of the model vary from −0.08to 0.01%, so it is decided to remove this variable from the model, which is leftwith 28 variables in total (four variables of ‘correlation’ were removed, one for eachdirection).Model tuning. We tested heuristically six different network configurations and exe-cuted each one 10 times to check its stability and trend. Table 1 shows the performanceof five models of ANNs, with different configurations, using the validation set. Thecalculus of the average F1-score reaches its maximum value when the ANN hasseven neurons in the first hidden layer and two neurons in the second one, so that wechoose this ANN for classification. The F1-score over the test set for the best modelwas 0.975.

4 G. Morales et al.

Table 1 Comparison between different ANNs

Type Architecture Performance(minimum)

F1-score overvalidation set(mean)

Hidden layer 1 Hidden layer 2

1 4 2 0.0122 0.9328

2 5 2 0.0103 0.9328

3 6 2 0.0108 0.9743

4 7 2 0.0100 0.9754

5 8 2 0.0094 0.9660

2.4 Classification

We first calculate BNDVI values of all pixels from each image after radiometric andatmospheric correction and set a threshold of 0.4 to distinguish between ‘non-clouds’and candidates to clouds (lower than 0.4). Then, for each pixel is calculated an Xdescriptor of 28 elements, which is processed in the neural network to decide if itcorresponds to cloud or not. Figure 2 shows the result of applying the describedclassification over four different scenes.

2.5 False-Positive Discard Method

Figure 2 shows an effective classification over those zones covered by clouds, butpresents some errors over some high-reflective buildings. Clouds shapes are morenatural and irregular, while building shapes have more regular shape outlines; thus,

Fig. 2 Cloud classification using the neural network. a, c, e, g Original images. b, d, f , h Detectedclouds in red

Cloud Detection for PERUSAT-1 Imagery Using Spectral … 5

we propose a method to exploit this information using a panchromatic fusion, whichincreases the spatial resolution and, therefore, the detail of the buildings.

We apply the ISH method, one of the fastest pansharpening algorithms [9], andtake only the blue component of the pansharpened image, as it is the componentwhere there is more contrast between high-reflective buildings and soil, as it wasobserved. Then, the first step is to analyze each small object classified by the ANNas a cloud, take as reference the pixel with the greatest intensity of the object (sothat we make sure not to take any pixel that is very near the edge of a building)and create a 50 × 50 window around it (Fig. 3a), which will be used to perform thepanchromatic fusion (Fig. 3c). Then, with this high-resolution image, we are able todetermine whether or not the central pixel is located on the rooftop of a building;that is, within a regular shaped object. To do this, the central pixel is taken as seedpoint to perform a simple region growing algorithm. The region is iteratively grownby comparing all unallocated neighboring pixels to the region (Fig. 3d).

To determine if the region surrounding the central pixel is regular, the Houghtransform is used over the edge mask of the region (Fig. 3e) to detect lines of at least13 pixels (9 m) as shown in Fig. 3f. If there is a straight line, the analyzed object willbe removed from the detected cloud mask. Finally, the resulting mask is dilated.

Fig. 3 Cloud classification using the neural network. a, c, e, g Original images. b, d, f , h Detectedclouds in red

6 G. Morales et al.

3 Results and Discussion

Figure 4 shows how the application of the explained method removes effectivelyregular shaped objects. The results for the 25 test images are shown in Table 2,where we obtained a minimum Kappa coefficient of 0.8065, a maximum of 0.9889,amean of 0.9430 and a standard deviation of 0.0513. The percentage of false positivesis significantly lower than false negatives. The false positives are mainly due to thepresence of high-reflective buildings whose regular edges could not be detected bythe proposed method either because of lack of contrast with their surroundings orbecause of a very small size. False positives are also due to the presence of high-reflective soil, which even for the human eye can be confused by clouds, and to thesea foam.

The Kappa coefficient is reduced mainly by the false negatives, which are mostlydue to the fog and cloud areas with low density, which in some cases border the largeclouds. Nevertheless, the proposed method proves to be able to recognize much ofthe areas affected by fog (Fig. 2f) and even clouds of very low density and smallsize. If the satellite images had a very different context than the images used for thiswork, the artificial neural network must be trained again to get similar good results.

4 Conclusion

Despite the different cases of false negatives, the results of cloud detection in high-resolution satellite images show high values of Kappa coefficients. In simple words,it means a high grade of concordance between cloud images selected by humanobserver and images detected by the proposed method. It is important to note thecloud coverage percentages of satellite images used to test vary from 0 to 90%approximately and not have relationship with Kappa coefficient or with the accuracy.

Fig. 4 Cloud classification using the neural network. a, c, e, g Original images. b, d, f , h Detectedclouds in red

Cloud Detection for PERUSAT-1 Imagery Using Spectral … 7

Table 2 Validation results

Image Kappa Accuracy Cloudcoverage(%)

Image Kappa Accuracy Cloudcoverage(%)

1 93.0324 99.87 1.04 14 96.4787 98.5 31.64

2 94.7298 99.96 0.44 15 94.276 97.61 30.8

3 97.6722 99.51 12.11 16 97.4476 98.79 39.17

4 97.4228 99.18 20.24 17 98.891 99.46 41.5

5 80.6516 96.08 13.36 18 97.3919 98.72 43.19

6 80.0462 91.32 35.82 19 97.8001 98.91 45.47

7 92.9886 97.9 19.34 20 96.2772 98.28 37.03

8 85.365 96.26 86.87 21 98.1937 99.21 32.61

9 93.2113 97.21 30.32 22 97.1358 98.57 49.79

10 94.6574 99.58 4.27 23 97.4612 98.73 53.28

11 90.0928 99.97 0.16 24 95.8069 97.9 52.32

12 97.9754 99.42 17.66 25 95.4834 98.97 13.56

13 97.1734 98.61 43.81

References

1. Tseng, D.C., Tseng, H.T., Chien, C.L.: Automatic cloud removal from multi-temporal spotimages. Appl. Math. Comput. 205(2), 584–600 (2008)

2. Hang, Y., Kim, B., Kim, Y., Lee, W.H.: Automatic cloud detection for high spatial resolutionmulti-temporal. Remote Sens. Lett. 5(7), 601–608 (2014)

3. Marais, I.V.Z., Du Preez, J.A., Steyn, W.H.: An optimal image transform for threshold-basedcloud detection. Int. J. Remote Sens. 32(6), 1713–1729 (2011)

4. Li, P., Dong, L., Xiao, H., Xu, M.: A cloud image detection method based on SVM vectormachine. Neurocomputing 169, 34–42 (2015)

5. Bai, T., et al.: Cloud detection for high-resolution satellite imagery using machine learning andmulti-feature fusion. Remote Sens. 8(9), 715 (2016)

6. Shi, M., et al.: Cloud detection of remote sensing images by deep learning. In: 2016 IEEE Inter-national Geoscience and Remote Sensing Symposium (IGARSS), pp. 701–704. IEEE, Beijing(2016)

7. Wang, F., et al.: New vegetation index and its application in estimating leaf area index of rice.Rice Sci. 14(3), 195–203 (2007)

8. Tsai, F., Chou,M.J.: Texture augmented analysis of high resolution satellite imagery in detectinginvasive plant species. J Chin. Inst. Eng. 29(4), 581–592 (2006)

9. Vivone, G., et al.: Critical comparison among pansharpening algorithms. IEEE Trans. Geosci.Remote Sens. 53(5), 2565–2586 (2015)

Implementation of a Digital ImageProcessing Algorithm on a Colibri IMX6Embedded Industrial System for OpticalMark Recognition

Carlos Herrera , Stephany Del Campo , Abel Dueñas , Julio León ,Guillermo Kemper and Christian del Carpio

Abstract This paper presents the implementation of digital image processing algo-rithms on Toradex’s Colibri IMX6 module and Iris carrier board. These algorithmswill be used for the extraction, detection, and recognition of optical marks on digi-tal images. Images will be obtained from an academic evaluation booklet through awebcam. Likewise, the computer will control an electrical/mechanical enclosure thatstores and transports the evaluation booklets to be registered. The system proposedresponds to the need to increase the efficiency of the grade recording process aspreviously designed for the Academic Coordination Department of the San Martínde Porres University’s School of Engineering and Architecture.

Keywords Linux · Embedded · Toradex · ARM · Images · BackpropagationPython · OpenCV · GTK+3 · Database

C. Herrera · S. Del Campo · G. Kemper · C. del Carpio (B)Faculty of Engineering and Architecture, School of Electronic Engineering, San Martín de PorresUniversity, Av. La Fontana 1250, La Molina, Lima, Perue-mail: [email protected]

C. Herrerae-mail: [email protected]

S. Del Campoe-mail: [email protected]

G. Kempere-mail: [email protected]

A. Dueñas · J. LeónSchool of Electrical and Computer Engineering (FEEC), University of Campinas, Av. AlbertEinstein – 400, Barão Geraldo, Campinas-SP, Brazile-mail: [email protected]

J. Leóne-mail: [email protected]

© Springer International Publishing AG, part of Springer Nature 2019Y. Iano et al. (eds.), Proceedings of the 3rd Brazilian Technology Symposium,https://doi.org/10.1007/978-3-319-93112-8_2

9

10 C. Herrera et al.

1 Introduction

At present, there are several options when it comes to embedded systems. One ofthe most used is the Raspberry Pi minicomputer, which has a Linux-based operat-ing system that is compatible with many software programs and a wide variety ofperipherals. The disadvantage of this embedded system is, however, that optimalperformance cannot be ensured if used for too long or with applications requiring ahigh level of processing. What is more, it has no protection measures for the systemand peripherals.

Another option in terms of embedded systems is offered by Toradex and its widerange of devices for industrial use, capable of giving comprehensive solutions thanksto their high performance, compactness, scalability, open-source embedded Linuxsystem development and compatibility with several peripherals. They are alreadyused in several industries such as the automotive, defense, laboratory, and measure-ment, e.g., the iGuide mapping system used to generate 3D virtual environments[1].

Due to all the features offered by the brand, it was decided to use Toradex’s ColibriIMX6 DL computer-on-module [3] that, along with its Iris carrier board, performswell in the implementation of digital image processing algorithms and hardwarecontrol.

2 Description of the System Proposed

2.1 Preparation of the Linux Embedded System

For the implementation of the enclosure control and processing system, it is requiredto choose the necessary elements. First, the operating system to be embedded mustbe run in the Colibri IMX6 module through the following steps:

• Create the Linux OS development environment (Ubuntu 14.04) on the computer.• Synchronize files from the “layers.openembedded.org” repository containing thelayers developed by Toradex for this module.

• Set up the device tree to add hardware.• Add the necessary compatible software to the system image configuration for theIMX6 module (main applications: Python 2.7 and OpenCV 3).

• Build the embedded Linux system for Toradex by cross-compilation with a GNUCompiler Collection (GCC) tool.

• Update the system on the computer-on-module.• Create an application development environment on the minicomputer [4].

Implementation of a Digital Image Processing Algorithm … 11

2.2 Detailed Description of the System

Figure 1 shows the pictorial block diagram of the integrated system, which consistsof a main component—the Colibri IMX6 module and the Iris carrier board—andfour secondary components—the graphical interface, the image capture device, themechanical enclosure and the MySQL database.

The main component, i.e., the Toradex single-plate computer (Fig. 2), was placedinside a transporting acrylic box. In addition, the processor was provided with amaxiFLOW heatsink [5] whose performance is better compared to that of normalheatsinks. Find below a list of peripherals used for the Iris carrier board [6] to connectto secondary devices.

(A) USB port to connect to the webcam.(B) RS232 connector (B) to connect to the mechanical enclosure.(C) Expansion port to touch control the LCD screen.(D) RJ45 Internet Port to connect to the database.(E) RS232 Port (A) to access the system console.(F) maxiFLOW heatsink.

For the implementation of image processing and recognition algorithms, thePython programming language [7] and the OpenCV image processing library wereused.

The optical marks were obtained from the evaluation booklet through a webcamera. Figure 3 shows the booklet’s yellow regions containing the optical marksthat students will use to enter their code and grade.

Figure 4 shows the enclosure where booklets are stored and transported througha mechanical system. Inside the enclosure, images of the booklet are also obtainedthrough a webcam. The handling of both the enclosure and the camera are carriedout with the Colibri IMX6 computer-on-module.

Fig. 1 Pictorial block diagram of the system developed

12 C. Herrera et al.

Fig. 2 Personalized single-plate Toradex computer

Fig. 3 Evaluation booklet to be processed

Implementation of a Digital Image Processing Algorithm … 13

Fig. 4 Mechanical enclosure for booklet storage and transport

2.3 Process for the Recognition of the Code and Grade on theBooklet

This project is based on a method that recognizes the code and grade on the booklets[2] using image processing techniques in Python, with the OpenCV computer visionlibrary [8] and the NumPy scientific library.

Figure 5 shows the steps taken in image processing to obtain the patterns requiredfor the trained neural network. First, a video capture object (C.1) is created usingthe camera ID number—in this case, 0. Thus, images are obtained from the camera,while setting up the parameters required, such as resolution (1024×576 pixels) andfocus.

captura = cv2.VideoCapture(ID) (C.1)

Then, the image of the object (captura) is imported by the read method, returningan 8-bit integer numeric matrix in the BGR color model, which is saved under frame(C.2).

_,frame = captura.read() (C.2)

At this point, the captured color image is ready, and the booklet, in this case,can be verified for the first time. To do so, the quantity of yellow pixels is to bedetermined.

First, the BGR color model (Blue, Green, Red) is converted to HSV (Hue, Satu-ration, Value) (Eqs. (1), (2) and (3) and C.3).

14 C. Herrera et al.

Fig. 5 Flow diagram of the recognition algorithm

V ← max(R,G, B) (1)

S ←{

V−min(R,G,B)

V I f V �� 0

0 otherwise(2)

H ←

⎧⎪⎨⎪⎩

60(G − B)/V − min(R,G, B) SiV � R

120 + 60(B − R)/V − min(R,G, B) SiV � G

240 + 60(R − G)/V − min(R,G, B) SiV � B

(3)

If: H < 0, then H � H + 360where R, G, and B are red, green, and blue pixel values, and H, S and V contain

values for hue, saturation and value.

img_hsv = cv2.cvtColor(imagen_bgr,cv2.COLOR_BGR2HSV)

(C.3)

where imagen_bgr is a copy of the original color image, and img_hsv is thenumeric matrix corresponding to the HSV image.

Then, a binarized image is created by thresholding (Eq. (4)), where lowerb andupperb are the lower bound and the upper bound corresponding to each pixel layer,and dst (I) is the resulting binary image. If the HSV pixel value is within the ranges,this will return the value 255; otherwise, it will be zero.

dst(I ) � lowerb(I )0 ≤ src(I )0 ≤ upperb(I )0 ∧ lowerb(I )1 ≤ src(I )1≤ upperb(I )1 ∧ lowerb(I )2 ≤ src(I )2 ≤ upperb(I )2 (4)

Implementation of a Digital Image Processing Algorithm … 15

Command C.4 performs Eq. (4) with the lower bound (nivel_bajo) and the upperbound (nivel_alto) and the resulting binary image (bin_amarillo).

bin_amarillo = cv2.inRange(img_hsv,nivel_bajo,

nivel_alto) (C.4)

The next step is to verify that the quantity of yellow pixels is within the range[25000, 50000]; for this, the value of all the matrix elements in the binarized image(bin_amarillo) is added and then divided by 255, as shown in Eq. (5).

cant_pixeles �∑1023

j�0

∑575i�0 pixeli, j

255(5)

After the first booklet verification, a cleanup of the binarized image(bin_amarillo) is done using command C.5 in Python, which will perform a mor-phological close operation with a 3×3 (8-bit integer) kernel.

bin_amarillo = cv2.morpholgyEx(bin_amarillo,

cv2.MORPH_CLOSE.kernel) (C.5)

In addition, objectswhose regions are greater than the threshold value are removedby segmentation. For such purpose, a find contours function is used, offering differenttypes of data, two of which will be used in this algorithm: the region of the objectand the centroid. In command C.6, the external retrieval mode (retr_external) andthe simple approximation type (chain_approx_simple) are established.

_,contornos,_= cv2.findContours(bin_amarillo.copy(),′′modo′′,′′aproximacion′′) (C.6)

Once the binary image of the matrix (bin_amarillo) is totally clean, all theobjects will be filled to obtain a clearly defined binary image with work regions(bin_regiones) so that an image of optical marks can be generated when executingthe “exclusive or” function (Eq. (6), C.7) between the binary image (bin_amarillo)and the image with work regions (bin_regiones).

dst(I ) � src1(I ) ⊕ src2(I ) (6)

where src1 (I) corresponds to the first binary image, src2 (I) to the second binaryimage, and dst (I) is the resulting image.

bin_marcas = cv2.bitwise_xor(bin_amarillo,

bin_regiones,mask = None) (C.7)

Then, a second booklet verification is made by counting the number of externalcontours (C.6), considering that the regions obtained must be equal to four (4): the