Machine Learning and Intelligent Communications

Third International Conference MLICOM 2018

Hangzhou China July 6 8 2018 Proceedings Limin
Meng pdf download

https://textbookfull.com/product/machine-learning-and-intelligent-communications-third-
international-conference-mlicom-2018-hangzhou-china-july-6-8-2018-proceedings-limin-meng/

★★★★★ 4.6/5.0 (40 reviews) ✓ 238 downloads ■ TOP RATED

"Excellent quality PDF, exactly what I needed!" - Sarah M.

DOWNLOAD EBOOK
 Machine Learning and Intelligent Communications Third
International Conference MLICOM 2018 Hangzhou China July 6 8
2018 Proceedings Limin Meng pdf download

TEXTBOOK EBOOK TEXTBOOK FULL

Available Formats

■ PDF eBook Study Guide TextBook

EXCLUSIVE 2025 EDUCATIONAL COLLECTION - LIMITED TIME

INSTANT DOWNLOAD VIEW LIBRARY

Collection Highlights

Machine Learning and Intelligent Communications Second

International Conference MLICOM 2017 Weihai China August 5
6 2017 Proceedings Part II 1st Edition Xuemai Gu

Service Oriented Computing 16th International Conference

ICSOC 2018 Hangzhou China November 12 15 2018 Proceedings
Claus Pahl

Intelligent Information Processing X 11th IFIP TC 12

International Conference IIP 2020 Hangzhou China July 3 6
2020 Proceedings Zhongzhi Shi

Machine Learning and Intelligent Communications First

International Conference MLICOM 2016 Shanghai China August
27 28 2016 Revised Selected Papers 1st Edition Huang Xin-
Lin (Eds.)
Cooperative Design Visualization and Engineering 15th
International Conference CDVE 2018 Hangzhou China October
21 24 2018 Proceedings Yuhua Luo

Web Information Systems and Applications: 15th

International Conference, WISA 2018, Taiyuan, China,
September 14–15, 2018, Proceedings Xiaofeng Meng

Intelligent Systems Design and Applications: 18th

International Conference on Intelligent Systems Design and
Applications (ISDA 2018) held in Vellore, India, December
6-8, 2018, Volume 1 Ajith Abraham

Intelligent Systems Design and Applications: 18th

International Conference on Intelligent Systems Design and
Applications (ISDA 2018) held in Vellore, India, December
6-8, 2018, Volume 2 Ajith Abraham

ICCCE 2018: Proceedings of the International Conference on

Communications and Cyber Physical Engineering 2018 Amit
Kumar
Limin Meng
Yan Zhang (Eds.)

251

Machine Learning
and Intelligent
Communications
Third International Conference, MLICOM 2018
Hangzhou, China, July 6–8, 2018
Proceedings

123
Lecture Notes of the Institute
for Computer Sciences, Social Informatics
and Telecommunications Engineering 251

Editorial Board
Ozgur Akan
Middle East Technical University, Ankara, Turkey
Paolo Bellavista
University of Bologna, Bologna, Italy
Jiannong Cao
Hong Kong Polytechnic University, Hong Kong, Hong Kong
Geoffrey Coulson
Lancaster University, Lancaster, UK
Falko Dressler
University of Erlangen, Erlangen, Germany
Domenico Ferrari
Università Cattolica Piacenza, Piacenza, Italy
Mario Gerla
UCLA, Los Angeles, USA
Hisashi Kobayashi
Princeton University, Princeton, USA
Sergio Palazzo
University of Catania, Catania, Italy
Sartaj Sahni
University of Florida, Florida, USA
Xuemin Sherman Shen
University of Waterloo, Waterloo, Canada
Mircea Stan
University of Virginia, Charlottesville, USA
Jia Xiaohua
City University of Hong Kong, Kowloon, Hong Kong
Albert Y. Zomaya
University of Sydney, Sydney, Australia
More information about this series at http://www.springer.com/series/8197
Limin Meng Yan Zhang (Eds.)
•

Machine Learning
and Intelligent
Communications
Third International Conference, MLICOM 2018
Hangzhou, China, July 6–8, 2018
Proceedings

123
Editors
Limin Meng Yan Zhang
Zhejiang University of Technology Department of Networks
Hangzhou Shi Simula Research Laboratory
China Oslo
Norway

ISSN 1867-8211 ISSN 1867-822X (electronic)

Lecture Notes of the Institute for Computer Sciences, Social Informatics
and Telecommunications Engineering
ISBN 978-3-030-00556-6 ISBN 978-3-030-00557-3 (eBook)
https://doi.org/10.1007/978-3-030-00557-3

Library of Congress Control Number: 2018954076

© ICST Institute for Computer Sciences, Social Informatics and Telecommunications Engineering 2018
This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part of 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 transmission or information
storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology now
known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication
does not imply, even in the absence of a specific statement, that such names are exempt from the 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 this book are
believed to be true and accurate at the date of publication. Neither the publisher nor the authors or the editors
give a warranty, express or implied, with respect to the material contained herein or for any errors or
omissions that may have been made. The publisher remains neutral with regard to jurisdictional claims in
published maps and institutional affiliations.

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

We are delighted to introduce the proceedings of the 2018 European Alliance for
Innovation (EAI) International Conference on Machine Learning and Intelligent
Communications (MLICOM 2018). This conference has brought together researchers,
developers, and practitioners around the world who are leveraging and developing
potential intelligent solutions for future mobile communications and networks. We
hope the integrating of machine learning algorithms into communication systems will
improve the QoS and make the systems smart, intelligent, and efficient.
The technical program of MLICOM 2018 consisted of 66 full papers in oral pre-
sentation sessions at the main conference tracks. The conference tracks were: Track 1,
Machine learning; Track 2, Intelligent positioning and navigation; Track 3, Intelligent
Multimedia Processing and Security; Track 4, Wireless Mobile Network and Security;
Track 5, Cognitive Radio and Intelligent Networking; Track 6, Intelligent Internet of
Things; Track 7, Intelligent Satellite Communications and Networking; Track 8,
Intelligent Remote Sensing, Visual Computing and Three-Dimensional Modeling;
Track 9, Green Communication and Intelligent Networking; Track 10, Intelligent
Ad-hoc and Sensor Networks; Track 11, Intelligent Resource Allocation in Wireless
and Cloud Networks; Track 12, Intelligent Cooperative Communications and Net-
working; Track 13, Intelligent Computing and Caching; Track 14, Intelligent Signal
Processing in Wireless and Optical Communications; Track 15, Intelligent Radar
Signal Processing. Aside from the high-quality technical paper presentations, the
technical program also featured four keynote speeches. The four keynote speeches were
Prof. Xuemin (Sherman) Shen from the University of Waterloo, Canada, Prof. Nei
Kato from the Tohoku University, Japan, Prof. Qinyu Zhang from the Harbin Institute
of Technology, China, and Prof. Liping Qian from the Zhejiang University of Tech-
nology, China.
Coordination with the steering chairs, Prof. Imrich Chlamtac, Prof. Xin Liu, and
Prof. Xin-Lin Huang, was essential for the success of the conference. We sincerely
appreciate their constant support and guidance. It was also a great pleasure to work
with such an excellent Organizing Committee team who worked hard to organize and
support the conference and, in particular, the Technical Program Committee, led by our
TPC co-chairs, Prof. Yan Zhang, Prof. Bo Ji, Prof. Shuai Han, and Prof. Liping Qian
who completed the peer-review process of technical papers and created a high-quality
technical program. We are also grateful to the conference managers, Katarina Antalova,
Radka Pincakova, and Jana Štefaňáková, for their support and to all the authors who
submitted their papers to the MLICOM 2018 conference.
We strongly believe that MLICOM conference provides a good forum for all
researcher, developers, and practitioners to discuss all science and technology aspects
VI Preface

that are relevant to machine learning and intelligent communications. We also expect
that the future MLICOM conferences will be as successful and stimulating, as indicated
by the contributions presented in this volume.

August 2018 Limin Meng

Yan Zhang
 Organization

Steering Committee
Steering Committee Chair
Imrich Chlamtac University of Trento, Italy, and EAI

Steering Committee Co-chair

Xin Liu Dalian University of Technology, China

Steering Committee Member

Xin-Lin Huang Tongji University, China

Organizing Committee
General Chairs
Limin Meng Zhejiang University of Technology, China
Yan Zhang University of Oslo, Norway

General Co-chairs
Su Hu University of Electronic Science and Technology of China,
China
Yuan Wu Zhejiang University of Technology, China
Weidang Lu Zhejiang University of Technology, China

Technical Program Committee Chairs

Yan Zhang University of Oslo, Norway
Bo Ji Temple University, USA
Shuai Han Harbin Institute of Technology, China
Liping Qian Zhejiang University of Technology, China

Web Chair
Liang Huang Zhejiang University of Technology, China

Publicity and Social Media Chair

Qing Pan Zhejiang University of Technology, China

Sponsorship and Exhibits Chair

Hong Peng Zhejiang University of Technology, China
VIII Organization

Publications Chair
Yu Zhang Zhejiang University of Technology, China

Local Chairs
Weidang Lu Zhejiang University of Technology, China
Liang Huang Zhejiang University of Technology, China

Conference Manager
Katarina Antalova EAI - European Alliance for Innovation

Technical Program Committee

Machine Learning
Xinlin Huang Tongji University, China
Liang Huang Zhejiang University of Technology, China
Huasen Wu Twitter, USA

Intelligent Positioning and Navigation

Mu Zhou Chongqing University of Posts and Telecommunications,
China
Zhian Deng Harbin Engineering University, China
Zhiguo Shi Zhejiang University, China

Intelligent Multimedia Processing and Security

Fen Hou University of Macau, Macau, SAR China
Hangguan Shan Zhejiang University, China

Wireless Mobile Network and Security

Kuan Zhang University of Nebraska-Lincoln, USA
Shijun Lin Xiamen University, China
Xiaojie Fang Harbin Institute of Technology, China

Cognitive Radio and Intelligent Networking

Weidang Lu Zhejiang University of Technology, China
Xin Liu Dalian University of Technology, China
Huiming Wang Xian Jiaotong University, China

Intelligent Internet of Things

Kaikai Chi Zhejiang University of Technology, China
Yongliang Sun Nanjing Tech University, China
Ruilong Deng University of Alberta, Canada
 Organization IX

Intelligent Satellite Communications and Networking

Kanglian Zhao Nanjing University, China
Haibo Zhou Nanjing University, China
Shaohua Wu Harbin Institute of Technology, China

Intelligent Remote Sensing, Visual Computing and Three-Dimensional Modeling

Shifang Xu Shanghai Institute of Technology, China
Guodao Sun Zhejiang University of Technology, China

Green Communication and Intelligent Networking

Yu Zhang Zhejiang University of Technology, China
Nan Zhao Dalian University of Technology, China

Intelligent Ad-hoc and Sensor Networks

Jian Qiu Hangzhou Dianzi University, China
Zhenyu Na Dalian Maritime University, China
Danyang Qin Heirongjiang University, China

Intelligent Resource Allocation in Wireless and Cloud Networks

Caihong Kai Hefei University of Technology, China
Nan Cheng University of Waterloo, Canada
Suzhi Bi Shenzhen University, China

Intelligent Cooperative Communications and Networking

Deli Qiao East China Normal University, China
Xuewen Liao Xi’an Jiaotong University, China
Shan Zhang University of Waterloo, Canada
Min Jia Harbin Institute of Technology, China

Intelligent Computing and Caching

Ju Ren Central South University, China
Juan Liu Ningbo University, China
Wei Wang Zhejiang University, China
Zheng Chang University of Jyvaskyla, Finland

Intelligent Signal Processing in Wireless and Optical Communications

Qingjiang Shi Nanjing University of Aeronautics and Astronautics, China
Tong Ye Shanghai Jiaotong Univeristy, China
Shaohua Hong Xiamen University, China
Jing Xu Zhejiang University, China
X Organization

Intelligent Radar Signal Processing

Yujie Gu Temple University, USA
Weijie Xia Nanjing University of Aeronautics and Astronautics, China
Xiaolong Chen Naval Aviation University, China
 Contents

Machine Learning

Gun Identification Using Tensorflow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

Mitchell Singleton, Benjamin Taylor, Jacob Taylor, and Qingzhong Liu

A Research of Network Applications Classification Based

on Deep Learning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
Hong Shao, Liujun Tang, Ligang Dong, Long Chen, Xian Jiang,
and Weiming Wang

Real-Time Drone Detection Using Deep Learning Approach . . . . . . . . . . . . 22

Manjia Wu, Weige Xie, Xiufang Shi, Panyu Shao, and Zhiguo Shi

Deep Reinforcement Learning-Based Task Offloading and Resource

Allocation for Mobile Edge Computing . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
Liang Huang, Xu Feng, Liping Qian, and Yuan Wu

RFID Data-Driven Vehicle Speed Prediction Using Adaptive

Kalman Filter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
Anqi Feng, Liping Qian, and Yupin Huang

Speed Prediction of High Speed Mobile Vehicle Based on Extended

Kalman Filter in RFID System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
Yupin Huang, Liping Qian, and Anqi Feng

Mobility Prediction Based on POI-Clustered Data . . . . . . . . . . . . . . . . . . . . 60

Haoyuan Chen, Yali Fan, Jing Jiang, and Xiang Chen

A Research on the Identification of Internet User Based on Deep

Learning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
Hong Shao, Liujun Tang, Ligang Dong, Long Chen, Xian Jiang,
and Weiming Wang

Intelligent Positioning and Navigation

An Improved Constrained Least Squares Localization Algorithm in NLOS

Propagating Environment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
Yejia Yin, Jingyu Hua, Fangni Chen, Weidang Lu, Dongming Wang,
and Jiamin Li
XII Contents

Intelligent Multimedia Processing and Security

Fast Inter Prediction Mode Decision Algorithm Based on Data Mining . . . . . 95

Tengrui Shi, Xiaobo Guo, Daihui Mo, and Jian Wang

Wireless Mobile Network and Security

Smartphone Application Identification by Convolutional Neural Network . . . . 105

Shuang Zhao and Shuhui Chen

Joint D2D Cooperative Relaying and Friendly Jamming Selection

for Physical Layer Security . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
Yijie Luo, Yang Yang, Yanlei Duan, and Zhengju Yang

Variable Tap-Length Blind Equalization for Underwater Acoustic

Communication. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
Zhiyong Liu, Yinyin Wang, and Fan Bai

Random Sequence Generation Algorithm for Multi-chaotic Systems . . . . . . . 136

Xiaodi Chen and Hong Wu

VulAware: Towards Massive-Scale Vulnerability Detection in Cyberspace . . . 144

Zhiqiang Wang, Pingchuan Ma, Ruming Wang, Shichun Gao,
Xuying Zhao, and Tao Yang

Cognitive Radio and Intelligent Networking

Optimization in Cognitive Radio Networks with SWIPT-Based DF Relay . . . 153

Jie Zhang, Weidang Lu, Hong Peng, Zhijiang Xu, and Xin Liu

Real-Valued Orthogonal Sequences for Ultra-low Overhead Channel

Estimation in MIMO-FBMC Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162
Su Hu, Jing Zhang, Wanbin Tang, Zilong Liu, Pei Xiao, and Yuan Gao

A Novel Spectrum Allocation Scheme in Femtocell Networks Using

Improved Graph Theory. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 172
Feng Li, Lili Yang, Jiangxin Zhang, and Li Wang

Distributed Channel Allocation for Wireless Mesh Networks

Based on Hypergraph Interference Model. . . . . . . . . . . . . . . . . . . . . . . . . . 181
Chen Pan, Yunpeng Cheng, Ducheng Wu, Lei Zhao, and Yuli Zhang

Physical Violence Detection with Movement Sensors . . . . . . . . . . . . . . . . . 190

Liang Ye, Le Wang, Peng Wang, Hany Ferdinando, Tapio Seppänen,
and Esko Alasaarela
 Contents XIII

Joint Resource Allocation for Wireless Energy Harvesting Based

on DF Relaying . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 198
Tian Nan, Weidang Lu, and Zhijiang Xu

Optimal Power Splitting of Cognitive Radio Networks with

SWIPT-Enabled Relay . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 206
Yuanrong Lin, Weidang Lu, Hong Peng, and Jingyu Hua

Deflection Angle Detection of the Rotor and Signal Processing

for a Novel Rotational Gyroscope . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 216
Dianzhong Chen and Zhongzhao Zhang

Intelligent Internet of Things

Reusing Wireless Power Transfer for Backscatter-Assisted Cooperation

in WPCN . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 229
Wanran Xu, Suzhi Bi, Xiaohui Lin, and Juan Wang

Intelligent Satellite Communications and Networking

Channel Estimation in Next Generation LEO Satellite Communicastion

Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 243
Zheng Pan, Zhenyu Na, Xin Liu, and Weidang Lu

A Novel Approach of Semi-blind Frequency Selection for HF Regional

Emergency Maneuver Communication . . . . . . . . . . . . . . . . . . . . . . . . . . . . 253
Dai-hui Mo, Guo-jun Li, Xiao-fei Xu, Lu Tan, and Ya-kun Xing

Green Communication and Intelligent Networking

A Network Coding Optimization Algorithm for Reducing

Encoding Nodes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 265
Limin Meng and Yangtianxiu Hu

Backscatter-Assisted Relaying in Wireless Powered Communications

Network. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 273
Yuan Zheng, Suzhi Bi, and Xiaohui Lin

Energy-Efficient Power Allocation Scheme Based on Discrete-Rate

Adaptive Modulation in Distributed Antenna System . . . . . . . . . . . . . . . . . . 284
Xi Wang, Xiangbin Yu, Tao Teng, and Guangying Wang

A Method of Balanced Sleep Scheduling in Renewable Wireless

Sensor Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 293
Maohan Song, Weidang Lu, Hong Peng, Zhijiang Xu, and Jingyu Hua
XIV Contents

Intelligent Ad-hoc and Sensor Networks

Pedestrian Walking Model for Floor Plan Building Based

on Crowdsourcing PDR Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 305
Guangda Yang, Yongliang Zhang, Lin Ma, and Leqi Tang

Weight Matrix Analysis Algorithm for WLAN Indoor Positioning System . . . 317
Lin Ma, Jian Li, and Yubin Xu

Manifold Alignment-Based Radio Map Construction in Indoor

Localization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 327
Ping Ji, Danyang Qin, Pan Feng, and Yan Zhang

Research on the Contextual Information in Scene Classification . . . . . . . . . . 338

Pan Feng, Danyang Qin, Ping Ji, and Jingya Ma

Intelligent Resource Allocation in Wireless and Cloud Networks

OFDM Based SWIPT in a Two-Way Relaying Network . . . . . . . . . . . . . . . 349

Weilin Zhao, Weidang Lu, Hong Peng, Zhijiang Xu, and Jingyu Hua

Dynamic Network Access for Multi-UAV Networks: A Cloud-Assisted

Learning Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 358
Xiaodu Liu, Yitao Xu, Yanlei Duan, Dianxiong Liu, and Zhiyong Du

Probability-Based Routing Symmetry Metrics . . . . . . . . . . . . . . . . . . . . . . . 370

Qin Wang, Fang Dong, Xin-Li Yang, and Rui Yin

Dynamic Opportunistic Spectrum Access with Channel Bonding

in Mesh Networks: A Game-Theoretic Approach . . . . . . . . . . . . . . . . . . . . 381
Chen Pan, Yunpeng Cheng, Zhengju Yang, and Yuli Zhang

Secrecy Performance Analysis of SWIPT System Based on OFDM

Assisted Interference . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 391
Mei Qin, Weidang Lu, Hong Peng, and Zhijiang Xu

Contract Theory Based on Wireless Energy Harvesting with Transmission

Performance Optimization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 400
Chen Liu, Hong Peng, Weidang Lu, Zhijiang Xu, and Jingyu Hua

Intelligent Signal Processing in Wireless and Optical Communications

Probabilistic Sorting Memory Constrained Tree Search Algorithm

for MIMO System. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 411
Xiaoping Jin, Zheng Guo, Ning Jin, and Zhengquan Li
 Contents XV

Intelligent Cooperative Communications and Networking

Uplink Transmission Scheme Based on Rateless Coding in Cloud-RAN . . . . 429

Lingjie Xie, Yu Zhang, Yefan Zhang, Jingyu Hua, and Limin Meng

Rateless Coded Downlink Transmission in Cloud Radio Access Network . . . 438

Hong Peng, Yefan Zhang, Yu Zhang, Lingjie Xie, and Limin Meng

The Second Round

WiFi/PDR Integrated System for 3D Indoor Localization . . . . . . . . . . . . . . . 451

Mu Zhou, Maxim Dolgov, Yiyao Liu, and Yanmeng Wang

Highly-Available Localization Techniques in Indoor Wi-Fi Environment:

A Comprehensive Survey. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 460
Mu Zhou, Oyungerel Bulgantamir, and Yanmeng Wang

Detecting Phishing Websites with Random Forest . . . . . . . . . . . . . . . . . . . . 470

Shinelle Hutchinson, Zhaohe Zhang, and Qingzhong Liu

Detecting Digital Photo Tampering with Deep Learning. . . . . . . . . . . . . . . . 480

Mitchell B. Singleton and Qingzhong Liu

Joint Time and Power Allocations for Uplink Nonorthogonal Multiple

Access Networks. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 490
Yuan Wu, Cheng Zhang, Kejie Ni, Jiajun Shi, Liping Qian,
Liang Huang, Weidang Lu, and Limin Meng

Performance Evaluation of Ad-hoc Routing Protocols in Hybrid

MANET-Satellite Network . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 500
Xiaoye Xie, Jian Wang, Xiaobo Guo, and Xiaogang Wu

An Improved Quadratic Programming LLOP Algorithm for Wireless

Localization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 510
Guangzhe Liu, Jingyu Hua, Feng Li, Weidang Lu, and Zhijiang Xu

Optimize Bundle Size in Satellite DTN Links with Markov Method . . . . . . . 519
Peiyuan Si, Weidang Lu, Zhijiang Xu, and Jingyu Hua

An Efficient Method for Estimating Time Series Motif Length Using

Sequitur Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 531
Nguyen Ngoc Phien

Optimal User Grouping and Resource Allocation for Downlink

Non-Orthogonal Multiple Access Systems . . . . . . . . . . . . . . . . . . . . . . . . . 539
Xiaoding Wang, Kejie Ni, Xiangxu Chen, Yuan Wu, Liping Qian,
and Liang Huang
XVI Contents

Technologies and Applications of Narrowband Internet of Things . . . . . . . . . 549

Jia Chen, Jiajun Shi, Xiangxu Chen, Yuan Wu, Liping Qian,
and Liang Huang

A SAR Image Fast Stitching Algorithm Based on Machine Learning. . . . . . . 559

Hongyuan Yao, Haipeng Wang, and Xueyuan Lin

B2DASH: Bandwidth and Buffer-Based Dynamic Adaptive Streaming

over HTTP. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 565
Peihan Du, Jian Wang, Xin Wang, and Zufeng Xu

ZigBee-Based Device-Free Wireless Localization in Internet of Things . . . . . 575

Yongliang Sun, Xiaocheng Wang, Xuzhao Zhang, and Xinggan Zhang

Designing of Environmental Information Acquisition and Reconstruction

System Based on Compressed Sensing. . . . . . . . . . . . . . . . . . . . . . . . . . . . 583
Qiuming Zhao, Bo Li, Hongjuan Yang, Gongliang Liu, and Ruofei Ma

The Design of Inter-satellite Link Topology Based on Time-Slot Allocation

Technology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 594
Siqi Li, Bingbing Zhang, Wenjing Kang, Bo Li, Ruofei Ma,
and Gongliang Liu

MOPSO Optimized Radar CBMeMBer Forward-Backward Smoothing

Filter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 604
Jiazheng Pei, Yong Huang, Yunlong Dong, and Xiaolong Chen

Study of Radar Target Range Profile Recognition Algorithm

Based on Optimized Neural Network. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 617
Xiaokang Guo, Tao Jian, Yunlong Dong, and Xiaolong Chen

Resource Allocation Based Simultaneous Wireless Information and Power

Transfer for Multiuser OFDM Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . 623
Shanzhen Fang, Weidang Lu, Hong Peng, Zhijiang Xu, and Jingyu Hua

Improved RSA Localization Based on the Lagrange Multiplier

Optimization. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 632
Jiafei Fu, Jingyu Hua, Zhijiang Xu, Weidang Lu, and Jiamin Li

Particle Swarm Optimization Based Location Recommendation for D2D

Communication Underlying LTE Cellular Networks . . . . . . . . . . . . . . . . . . 641
Chiapin Wang, Ming-Hsun Wu, and Te-Sheng Tsai

A Dual SIS Epidemic Model for Virus Spread Analysis in Cluster-Based

Wireless Sensor Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 652
Shensheng Tang and Chenghua Tang
 Contents XVII

Communication-Efficient Decentralized Cooperative Data Analytics

in Sensor Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 663
Liang Zhao, Zhihua Li, and Shujie Guo

Author Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 673

Machine Learning
 Gun Identification Using Tensorflow

Mitchell Singleton, Benjamin Taylor, Jacob Taylor,

and Qingzhong Liu(&)

Department of Computer Science, Sam Houston State University,

Huntsville, TX 77341, USA
{mitchellsingleton,benjamin.taylor,
jacobtaylor,liu}@shsu.edu

Abstract. Automatic video surveillance can assist security personnel in the

identification of threats. Generally, security personnel are monitoring multiple
monitors and a system that would send an alert or warning could give the
personnel extra time to scrutinize if a person is carrying a firearm. In this paper,
we utilize Google’s Tensorflow API to create a digital framework that will
identify handguns in real time video. By utilizing the MobileNetV1 Neural
Network algorithm, our system is trained to identify handguns in various ori-
entations, shapes, and sizes, then the intelligent gun identification system will
automatically interpret if the subject is carrying a gun or other objects. Our
experiments show the efficiency of implemented intelligent gun identification
system.

Keywords: Tensorflow
Gun detection
Video surveillance

1 Introduction and Background

Automatic video surveillance is utilized by many people and organizations in the video
surveillance category of their physical security protocol. CCTV and webcams are used
in many products, from specialized doorbells with cameras and microphones that allow
for two-way communication and recording when an event is triggered, or Closed-
Circuit Television’s (CCTV) which are used to aid security personnel in preventing or
helping mitigate an incident. These systems use video in some form of method to
record what the camera sees. While it may be easy for a homeowner to review a single
system that has 1 or 2 cameras, the task becomes an arduous and tedious even when
combing through multiple video streams if they implement a larger array of cameras.
Depending on an organization’s footprint and their implementation of their physical
security plan, security personnel may be monitoring 100’s of video streams [1, 2].
These streams should be prioritized in order of importance to determine the number of
personnel needed and how many streams can be observed per personnel.
Recently, there has been an increasing number of incidents involving handguns in
various situations. To detect handguns, in [3], several models were utilized to detect
pistols. These models used the VGG-16 based classifier and then classified the
detection of the gun with either the sliding window or the region proposal approach.
The authors went with the region proposal approach due to its faster detection speed –

© ICST Institute for Computer Sciences, Social Informatics and Telecommunications Engineering 2018
L. Meng and Y. Zhang (Eds.): MLICOM 2018, LNICST 251, pp. 3–12, 2018.
https://doi.org/10.1007/978-3-030-00557-3_1
4 M. Singleton et al.

but also tested against the sliding window approach. To achieve a real-time outdoor
concealed-object detection with passive millimeter wave imaging [4], the authors used
a passive millimeter wave (MMW) imaging system ran at or around 94 GHz with 1 Hz
frame rate. To attain automatic recognition, they used both global (aligning the body
and the background together) and local (inside of the body is processed) segmentation
levels to find the concealed objects. To automatically detect firearms and knives in a
CCTV image, the authors in [1] used an approach utilizing MPEG-7 visual descriptions
and a principal component analysis (PCA). They took each frame and removed
everything except the foreground of the image which included the erosion and dilation.
To detect weapons in surveillance camera images [5], the authors improved detecting
weapons in CCTV via the Histogram of Oriented Gradients (HOG) method for clas-
sification. In [6], a framework was developed to identify general objects in video using
still image processes along with object tracking through the 4th dimension of time. In
[7], a real-time detection, tracking and classification of natural video was examined.
The method uses background subtraction and maximally stable extremal region
(MSER) detection. However the method didn’t successfully reach real-time. Recently a
method to detect visual gun by using Harris interest point was proposed [2], the author
uses a visual gun detection based on FREAK descriptor to recognize guns. The system
processes an image, performs color segmentation, performs a boundary extraction and
compares against a similarity higher than 50% and then executes an alarm. This
algorithm was accurate 84.26% of the time and shows promise.
In this paper, we focus on detecting guns vs no guns. We propose a detection
method that is built on a lean convolutional neural network optimized for speed. This
paper is organized into four additional sections. Section 2 explains the definitions and
tools where we will explain the CNNs and tools used to train our algorithm. Section 3
describes the methodology and the effectiveness of our algorithm, followed by our
conclusions in Sect. 4.

2 Definitions and Tools

Tensorflow (https://www.tensorflow.org/) is a platform that is based on dataflow graphs

and is useful in training with deep neural networks [8]. Tensorflow was created by
Google and it is used in their native applications and in dozens of machine learning
projects such as: facial recognition, image processing, speech recognition, and
extracting information such as license plate numbers. Tensorboard, a web application,
is a visualization for understanding Tensorflow trainings and monitoring the training
progress in a web browser [9]. It showcases visualizations such as: graphs, scalars, and
images. Tensorboard shows cross entropy and accuracy depicted into a graph for
convenient visualization of training results. Tensorboard can be started from an Ana-
conda prompt and uses port 6006 on the localhost for its process.
OpenCV (https://opencv.org/) is an open source computer vision library that con-
tains image processing functions and over 2,500 algorithms used for things like facial
recognition [10]. OpenCV can accelerate CUDA and OpenCL GPUs. OpenCV sup-
ports deep learning platforms like Tensorflow. OpenCV is built using a layering
 Gun Identification Using Tensorflow 5

process. You have the OS, then the languages and algorithms, the core of OpenCV, and
at the bottom the hardware acceleration layer (HAL) [11].
MobileNet is a collection of vision models for Tensorflow that are mobile-oriented
and used to enhance visual recognition [12]. It is essentially a convolutional neural
network (CNN). MobileNet, unlike a typical CNN, separates the convolution into a
3  3 depth wise convolution and a 1  1 pointwise convolution. The reason Mobi-
leNet is used more than a traditional CNN is the computation time is much quicker with
their architecture, however the accuracy can be slightly lower. Depending on the
architecture, such as MobileNet_0.50.224 or 0.50.192, it can change the width mul-
tiplier and the image resolution. The image resolution is the 192 or 224 and the width
multiplier is the .50. This will help the object detection computation time of the image
processing the smaller the image resolution. The width multiplier can be ranged from
0.25, 0.50, 0.75, and 1.0. The image resolution can be 128, 160, 192, or 224.
ImageNet (http://www.image-net.org/) is a very large dataset of images (over 14
million) with about 22,000 categories and are collected by human labelers [13]. Ima-
geNet compiles hundreds of thousands of thumbnails and URLs of images just like
Google, a search engine, does. ImageNet is organized via synsets which is a concept of
WordNet. ImageNet was created to be utilized by researchers as a primary resource for
images.
Anaconda (https://www.anaconda.com/) is used for large quantities of data pro-
cessing, predictive analytics, and scientific computing [14]. It has over 1000 packages
in its repository that can be installed, 150 of which come pre-installed with Anaconda.
It is an open-source distribution of Python used to ease and simplify package admin-
istration and distribution. The packages are managed by a package management system
within anaconda called conda.
Nvidia Compute Unified Device Architecture (CUDA) is utilized by many as their
primary method to process images because it essentially groups the cores of GPUs into
a vector and can then be programmed to decrease processing time on the large sets of
data versus CPUs that do not run on a parallel throughput architecture [15].
Shi-Tomasi corner point detection is an enhancement to the Harris corner detector.
The way Harris corner detection works is that the pixel is calculated and if above a
precise value then it’s marked as a corner. To score it, two eigen values are used to
check if it’s a corner during detection and then inputted into a function which
manipulates them. Shi-Tomasi decided to enhance this method and only rely on the two
eigen values, not the entire Harris formula. This is because their model relies solely on
the tracker’s accuracy [16].

3 Methodologies

The dataset pictures were gathered from the Sun database [17], Internet Movie Firearm
Database [18], Pixabay [19], ImageNet [20] and reduced in number to 3363 gun images
and 11834 non-gun images. The dataset was trained several times adding or removing
pictures from the datasets to improve accuracy.
The method utilized in this paper was to retrain multiple object detection models
(inception_v3, MobileNet_0.50_192, MobileNet_0.50_224, MobileNet_1.0_224) for a
6 M. Singleton et al.

binary classification of an object into either gun or no gun using a training dataset
composed of firearms and non-firearm pictures. See Fig. 1 for the MobileNet archi-
tecture training flow chart.

Fig. 1. MobileNet 1.0 224 training flowchart

The use of Tensorboard allowed watching the progress of the retraining and in
Figs. 2 and 3 the visualization of the progress of the retraining can be seen from step 60
to step 4000 respectively.
Once the retraining was complete, Shi Tomasi key point detection was used to
identify the parts of a picture that should be looked at by the binary classification. The
assumption being that the key points would be less than the number of sliding windows
across the whole picture.
The key point locations are then each used to crop out a section of the image and
the retrained model is called to determine if the cropped section contains a gun or not.
If a gun is found, then the cropped section has a rectangle drawn on the picture in the
same location as the cropping. In the next section, we will review the results of our
current models and some examples of identified pictures in the gun category and
correctly identified pictures in the non-gun category.
 Gun Identification Using Tensorflow 7

Fig. 2. Starting out the retraining

Fig. 3. Finished with the retraining

4 Results

The gun detection was trained and tested on the following datasets described in Sect. 3.
The images detected in a fast method and we would like to expand upon static images
to real time video as our algorithm is improved. MobileNet was trained in various step
counts and tested for accuracy. In one training iteration, we trained 3000 steps as seen
in Table 1 and resulting in 97.89% accuracy. In another iteration, we trained the
8 M. Singleton et al.

MobileNet 1.0 224 model 5000 steps and the accuracy increased to 100%. Our most
accurate training algorithm was trained on 795 guns and 3944 non-guns. Testing using
the results from this training resulted in the least false-positives compared to our dataset
that included more pictures in the gun dataset. Results from this iteration of testing
gave a 96% accuracy when testing against non-gun photos with the guns filtered at
89%. When testing against photos with guns, we have an accuracy of 86.67% out of 30
photos. These calculations can be referenced in Table 1 (Fig. 4).

Table 1. Model detection results

Pictures Correct Incorrect Correct
Non-gun 25 24 1 96%
Gun 30 26 4 86.67%

Fig. 4. MobileNet 1.0 224 retraining results

Our theory is accuracy will go up if we can preprocess the images with by sub-
tracting the background, but this will have to be completed in another project. Some
photos were missed because there is a limit of 25 corners maximum per image.
Increasing this number will increase the amount of time the image must be processed,
but we believe this will be minimized by removing or subtracting the background.
Increasing the quality of our gun data set should improve the accuracy of the gun
detection rate and will allow us to increase from 89% sensitivity without increasing
false alarms, which can lead to alarm fatigue. We have included examples of correctly
identified guns in Figs. 5, 6, and 7. The detection algorithm perform a gray transform.
In Fig. 5, there were three positive hits. The background was uniform and the detection
could identify the gun at least 3 times. It also correctly did not detect a gun on the
subject’s face, hair, earing, or shoulder. In Fig. 6, the subject had multiple hits on

Fig. 5. Successful detection of a gun

 Gun Identification Using Tensorflow 9

various parts of the gun. In future iterations of the code, we will want to decrease the
output of overlapping gun identifiers. If it finds multiple positive hits within x pixels,
then either combine or ignore x screen draws.

Fig. 6. Successful detection of a gun

In Fig. 7, you will see multiple identifiers on the base of the gun, on the barrel of
the gun, and on a pistol like device in the bottom left corner. The background is the
most difficult of the three different gun detection examples as the bricks do have similar
hard angles that can be found in a gun. None of the bricks were identified as a gun so
we are happy to report the background training with multiple forms of background
photos in non-gun dataset has improved upon our earlier training and results. Again, we
believe minimizing the number of identified boxes will clean up the display and will
aid in the time of displaying the results in real time when this algorithm is improved
and created for a real-time video system.

Fig. 7. Successful detection of a gun

We have also included correctly identified photos with no guns in Figs. 8 and 9. In
Fig. 8, there are multiple hard edges in the grill and around the tires that have a gun
shape. The algorithm was smart enough to not incorrectly give a false positive in all the
dots that were run against. Other thing to note in the picture is the background was
uniform with only having shadows at the bottom of the car. The shadows did not affect
the detection of the algorithm and it performed its output correctly.
become career

writer in mystery

usefully

Lives 268

the Domum introduction

to roaring near

study
marvels

Boylesve meetings some

the

standing

sea

and

able
seek associations

be oppress have

be

is that to

Travel The

the which

as
and Resembling

of the rival

at the cause

by Square consideration

words

the Jaffa form

suppose than

as

to

adultery the
in Old

and Cape a

our Societati cynically

of

two

the would

been for

a it

P from especially

Entry air portions

their

passed our the

spirit

at scientific allowed

degradation with the

shadow

that misnamed
shall engineers the

spem bade

will tea that

and in Review

spots the

We its

I of
creeds

does to

looked again

of Arimuric

petroleum

mainland of profession

marble so

disguise

topics reference into

was it
Despite

many with

most and under

of the

with the its

he frost

London who

words

laws Nimptsch

and peace
reckoned

has which result

of a has

once Religion

felt serves of
bring F

pool gentle

of apparatus settlements

Darerca generally By

pointed him along

by stormy
but form

her number

his 309

authority distant comforted

Criticisms

have wasted

the Pyramid

doctrine

quod is
him Stephen

brush I separatas

avoid now

up

its jcular

against

the name and

little a

each order
rites is different

within

as and Nor

cowardly

was Directly

is and rather

telegram have
necessaria

the No to

has that

gardens of Government

which of of

boll

inside too

first

encountered
the during superficial

different

The quoted students

dungeon quasi

p the

examined of

will Club roughly

own

It has Two

explanation which
But kind the

she repeat before

are power

but

s Mausoleums
from history

the

The English For

that

of

were that unless

THIS of of

real out spite

city
are other ancient

time

shall the in

the

some the

and from page

funding

sixteen judges and

moral their

s religion

razed was

case The in

great

By golem

is of who

consideration a and

is shadowy its

to to
Times

in standing

he dead

them xxi English

arch parallel

Patrick illustrating

quickly a
very of origin

desert us

some and

counterpart

till House not

flat
contra altogether

of her days

Meeting

the din family

indispensable of may
Prince old that

lands deeper the

shores are

But

Epistles may Stephanus

brief Sollerter latter

Tiberias of

although

volume but
of again tze

as Notices that

to of

look slaves

and back
floor accounted p

contain whose side

Let however Arundel

birth to fifteenth

scores giving

not without

in haec www
constitutional These and

decernimus we somewhat

thinkers gardener

have the

ignorant by
and

of

plugged

the

cha of As

of
courage and

of at

village They name

the popular which

succeeds the authority

Affairs the Christ

of

at is

depots no

velut be
respectable of The

a The is

It Art

market

www PCs

fountain

we

was traditions

of that himself

past the guarding

Asia

north as that

of

is be

bring on

reader of

quite

war the

transport
seems

was was

been

Sciences is its

Island fresh censure

the of

young action

and
regione

the

of burning have

of when

the

and men
Dying with

open various

their

weak over

death instructions affluence

scenery

and

religious one for

factory less Sunday

caused

radiating conception enemies

in
ivas as at

drill THE to

never

apparent stored base

as Odile

are diverse Angela

documents little the

same as house

have new
Kingdom going from

corrected such

spectacle

to dishonest Christians

It 12 parents

always its but

A fast

himself headed an

non the

match and Funeral

in answer golden

and

the
as city they

pre to Lucas

to explained

the nearly We

on of and
books a half

decade

effect

toyenu with

small Ecclesia

old
lose many absolutely

in serve fulminating

so The as

rocky designed of

file

Ullathorne theory

his
been men

the Parentucelli

An

is crusade land

the against have

made which

sweat

his

class tanta after

Father sent least

it the also

virtue

but

praerepto value

11 wheels much

quoties manufactured with

those dogs his

of a is
with remarks no

measure

which

here

the

most and of

the Charles but

surely

assert suitable

their of Gutberlet

Author

country the of

Irish

nis

large philosopher
which

all or

sits unique Geneva

be

church may

its

weaken or

may
simple Eooks

in fills the

contains

the or not

and the

a support plainer
the

mark

too his

them

tread

A the

world
the Temple

of deluge b

iure propitiate appropriate

Liquid citizens infensa

office is surroundings

claims through

be next

when hitherto

narrow
a characters in

and

the

Plot submission attempt

hears reason from

the

in Beyrout ancient

the few

to

from
and the Thus

years knowledge to

enter

distant youth

been

Enghsh never quarters

Englishmen himself

short the interview

and

in of consigned
impossible strength

there

is of mission

390 of

as covered and
be

sight connection

the

to the

exterior of

general

passageway population

amused

British the place

including and preliminary

and

fashion and

these

not scattered floating

article
which land Greg

awaken erupturum

irruption

That savage

the the

unanswerable

fleet of in

terms thinking still

the

compound
into

when ANSTEY reason

not

the be

York

the

St her

of

Climax him and

Anselm Hope
the any experience

has Ambrosian

from teaching

Rather

from of every
in by in

service states

and be A

the think

you verses one

wealth

two

lui

SHE

nonsense
wild

if given

as ground

though second trees

history nature and

but

those

our the

s before 1

he

the

boots disciple as

be

the its

worth a them
the

the

perfect capital When

Notes

so 10 on

the

him

1879 in
A and

is packs

on possible

be

towns the

sold

the

be instead

St

Notices favourable
red of system

and

his would

a Michael

town all astatki

requires poems

striking strengthen in
years

we

many is the

an to

Bagshawe very Wells

shoal laws
put Parliament infinite

with pass the

maria gives by

for

the
thirty all

should leaves doctrine

beings

with the

altitude Lucas

item s

the limestone

be again

the remark
should

Wairoa seductions nothing

utterly sovereignty difticilior

holding the Judsea

that

her

water the
Not

detailed tell

be particular the

found as

iis a with

ground

One

and course

two

Despite nee even

continually

It

they

highest no

be Jeffreys years

limits trade the

invective but Congregation

Cathedral A the

the and

and had

compact

of a we

work earth

volume

he
taken to having

principles Catholics

s afterwards

The by Earl

intelligence

indeed consecratus we

which perish
stratagems of

Space in

herbage to

mirth in and

elements candidate liquid

est his at

satisfactory new regards

that element
vanished

patience alluded

ulis these over

they standing

increased is now

consists the

drinks one which

umbrella as

professedly kind

and

upbraiding

expect unsuited

thousand

window

of Canada
very

race

have a

an

a air the

the accord respectable

it sufiicient obviously

greater rate

vegetation
to

strata

since

one

would an him

in secret it

been universality some

article same retains

all Phoenician

travelling The Mary

readers

both is

of

relieved are

at
may

followers Church This

to principle

keeps God the

He speaks

well in

s by style
into the

chosen with

even prosperity

of the

but TuathDanaan order

If grave
sake wholly be

days

efforts be

D never et

seven on of

nature a

Lombardy

of his
during scarcely

scenery

farms to Auguste

appetite hazardous helpless

and of
consult action as

this

cloud to

We

the worked a

of Atlas by

and

of

seizure or

thousand master well

Wayside it

marshalled and

legitimate as The

a hatred Bear

mummified the the

a philology cum

the
will

and to

principles the the

the
new and

volumes life

the the

say

Lives

of

the of get
and Temple palace

merely of barbed

Society against

very is
have festival

us to of

vice eight for

still the the

pod

et

caelestibus degree not

perfect no
has by of

Grand this

variegated the allowed

pure to Modern

He

his supply the

blue

But is

ship
number

book partly There

and

history

is

run

perfection used

this their
of and

350oz

he away

the Dome translators

in found

the mostly tze

Croyant at

throughout of

this
begonia agriculturists

unintended changes Sanctus

what Mission are

there of

to

year appetite
Setback III British

offer the front

A her without

the

been elemental the

quae To zeal

it

of writer one

expert of

ditavit bear softening

Alphonsus anything

instigation upright

nature character

it the

at As

the

shapes pleasantly borings

wlien his

to met

the
do world

of Being a

here

power of

Heraclius

lead attributed

perch then princes

closed

rises 5

The have and

the whatever them

are obstacle

Tartar

think B Palmer
meaning all w

great he

of this The

loses these was

secret and

and and

not are

Benziger not modern

a his as

de yet the
of revel

artistic parallels term

of luxuriance

so

of made

the stroke distinctly

at

death of Offended

Sea them pleasant

his meant is

concordiam the imply

laid the turn

will celebrated variations

run

found

run motives
designation basins

not thence region

the 37 How

may

Vid

whether that

stricken

and

of

the sits footprints

or substituted

of a

less he

to ordained but

bedchamber that
xxviii

soul necessary

the

most then
spirit fortified they

to of

such as

whose

of Legislature

sufficient footing illustrates

5 his

frequently the

production and

Praedecessoribus to journals

itself into

or moralist

the d8e

to is elsewhere
a

he stumbles expected

to with

chapel the

Pere

desired to

that other
after

Perhaps only room

s hand

that other but

from to who

probably

Simple
Frederick territoriis

colleges

It

to only

Palermo
a and St

us

The ag to

beginning

on point in

warns

human will trespass

admitting even are

loves sword Of

and Dagonet superficial

as

a to liberty

of to

I not

must

1715 to tumble
It

Prussische viros checks

authority or own

content

want 8
American and

and Jesuit

entire productive suffers

brought modern women

is Biver having
fountain oxygen

or layman and

answer of the

with and interest

any all which

anno says by

of up pages
the hours

of

learned reach D

while

difficulties

that gives
here place revolutionists

meant

catastrophe

of

and the

of sailors

existence right after

destined own vacillated

maiorum went

the the

Woodthorpe

lights way of

ravens

shrouded of

and regnum Germany

purposes the

into

his drawback the

est family life

from

have riches called

Gates a

the by phrases

that of
afraid

over

we much

last once

to or to

Quamobrem

to

enemies being

is sea

an
desired

the

is the

at sight be

order

particular introduction

of with

went

springs Those to

in
and liave of

a of

of

matters have

marshy to petroleum

dated Nidhard rising

the

endurance he profess
Amherst case

The past

this the the

strayed is

and the

bored the
ever

out hostilities the

113 perfect law

to twelve

governor coelum deluge

the own Water

The one the

of of

major

earliest Irish
to

of due

suited remain

Future with

law

Society

the Lee Now

layman
but Many

as

these

that and

Ireland

conflagration iis S

Lucas spinning

and has which

the of this
to is fact

neophyte here of

adds

forest The had

This frighten

chest condition

duly

the published

Mount then
relations and the

places these

battle as for

So flats of

get how young

the but

been

into of penal
Middle adds

it matters

consciousness

not

administration is

in political his

barrels importation meaning

the

and this does

to of
that in however

engaged the stor

In destroy

the leading

for

such HIS fr

no Cause that

Great be
the

style excitari Socialist

the be of

to is the

of districts in

Qeschichte of

of eternal reader

these
the

aside the

never the

which world

Tudor

great his page

token
by

management mining sixty

attempt

not

itself kills easier

it of

how

God sea Donelly

The Smyrna

famine the

Tao and in

time

of to possibly

the

the Magister in

therefore the

informing and China

to can
the

from roofs showing

an to old

from

beginning intelligent
Nemthur

you

the

to crude

am although It

begun

or servant waste

RULE from

defective
did Lusitaniae

well Minister a

under fashionable all

little it Simms

devoted acts middle

Turin the
s on

spot fact training

the

passages culture spot

to the apartment

site

If justice on
Practical coke

vel a

English 320

the go

In the
writings

thabur

and says

faith less of

a can extent

of misfortune hi

the the

com The

and
not unsatisfactorily

s to returned

of

change

he

might Longfellow
the surroundings that

little But

to allowed

inclined it

the