Film

MIGRATE_ELM.pdf

Description
MIGRATE_ELM.pdf
Categories
Published
of 23
All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.
Related Documents
Share
Transcript
  Full Terms & Conditions of access and use can be found athttp://www.tandfonline.com/action/journalInformation?journalCode=tjit20  Journal of Information and Telecommunication ISSN: 2475-1839 (Print) 2475-1847 (Online) Journal homepage: http://www.tandfonline.com/loi/tjit20 Extreme deep learning in biosecurity: the case of machine hearing for marine species identification Konstantinos Demertzis, Lazaros S. Iliadis & Vardis-Dimitris Anezakis To cite this article:  Konstantinos Demertzis, Lazaros S. Iliadis & Vardis-Dimitris Anezakis (2018):Extreme deep learning in biosecurity: the case of machine hearing for marine species identification,Journal of Information and Telecommunication, DOI: 10.1080/24751839.2018.1501542 To link to this article: https://doi.org/10.1080/24751839.2018.1501542 © 2018 The Author(s). Published by InformaUK Limited, trading as Taylor & FrancisGroupPublished online: 02 Aug 2018.Submit your article to this journal Article views: 23View Crossmark data  Extreme deep learning in biosecurity: the case of machinehearing for marine species identi 󿬁 cation Konstantinos Demertzis a , Lazaros S. Iliadis b and Vardis-Dimitris Anezakis  c a School of Engineering, Department of Civil Engineering, Lab of Mathematics and Informatics, DemocritusUniversity of Thrace, Xanthi, Greece;  b School of Engineering, Department of Civil Engineering, Lab of Mathematics and Informatics, Democritus University of Thrace, Xanthi, Greece;  c Department of Forestry andManagement of the Environment and Natural Resources, Democritus University of Thrace, Orestiada, Greece ABSTRACT Biosafety is de 󿬁 ned as a set of preventive measures aimed atreducing the risk of infectious diseases ’  spread to crops andanimals, by providing quarantine pesticides. Prolonged andsustained overheating of the sea, creates signi 󿬁 cant habitat losses,resulting in the proliferation and spread of invasive species, whichinvade foreign areas typically seeking colder climate. This is oneof the most important modern threats to marine biosafety. Theresearch e ff  ort presented herein, proposes an innovative approachfor Marine Species Identi 󿬁 cation, by employing an advancedintelligent Machine Hearing Framework (MHF). The  󿬁 nal target isthe identi 󿬁 cation of invasive alien species (IAS) based on thesounds they produce. This classi 󿬁 cation attempt, can providesigni 󿬁 cant aid towards the protection of biodiversity, and canachieve overall regional biosecurity. Hearing recognition isperformed by using the Online Sequential Multilayer GraphRegularized Extreme Learning Machine Autoencoder(MIGRATE_ELM). The MIGRATE_ELM uses an innovative DeepLearning algorithm (DELE) that is applied for the  󿬁 rst time for theabove purpose. The assignment of the corresponding class  ‘ native ’ or  ‘ invasive ’  in its locality, is carried out by an equally innovativeapproach entitled  ‘ Geo Location Country Based Service ’  that hasbeen proposed by our research team. ARTICLE HISTORY Received 30 November 2017Accepted 14 July 2018 KEYWORDS Biosecurity; invasive species;machine hearing;geolocation-based services;extreme learning machines;big data 1. Introduction 1.1. Bio-safety and bio-pollution Bio-safety (BISA) is a strategic and integrated approach. It includes policy and regulatoryframeworks, plus risk analysis and management rules. It is related to food safety, animaland plant life and health, in certain environments. It covers the introduction of plantpests, animal pests, diseases and zoonoses, and the release of genetically modi 󿬁 ed organ-isms. BISA is a hot topic that has emerged recently is the introduction-management of Invasive Alien Species and their genotypes. It is essentially a holistic approach that is © 2018 The Author(s). Published by Informa UK Limited, trading as Taylor & Francis GroupThis is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/ licenses/by/4.0/ ), which permits unrestricted use, distribution, and reproduction in any medium, provided the srcinal work is properly cited. CONTACT  Vardis-Dimitris Anezakis danezaki@fmenr.duth.gr Department of Forestry and Management of theEnvironment and Natural Resources, Democritus University of Thrace, Pantazidou 193, PC 68200 Orestiada, Greece JOURNAL OF INFORMATION AND TELECOMMUNICATIONhttps://doi.org/10.1080/24751839.2018.1501542  directly related to sustainability in areas such as agriculture and food safety and it refers tothe protection of the environment, mainly focusing in biodiversity. The threats to biosecur-ity, are related to small-scale risks that emerge rapidly, making the application of ane ff  ective policy, a very challenging task. This is due to time and resources ’  limitations,which make the analysis and assessment of threats likelihood, a tedious task.IAS are a result of generalized climate change and they constitute a serious and rapidlyworsening threat to natural biodiversity in Europe. They enter new foreign habitats andthey can sti 󿬂 e natural  󿬂 ora or fauna, causing serious harm to the environment (Demertzis& Iliadis, 2017a). The impacts are socio-economic and there are negative consequences inpublic health, in  󿬁 sheries, in agriculture and more generally in food production. The inva-sion of these species is the second biggest threat to local biodiversity worldwide and iscalled  ‘ bio-pollution ’ . The impact of IAS and their rapid expansion in new seas havedirect economic consequences in various areas (e.g.  󿬁 nancial) such as the risk of indigen-ous species ’  extinction, entails costs to restore natural balance. The expansion of organ-isms harmful to the human health might lead (in mid-term) in the reduction of touristdevelopment. Ecological impacts such as food grid disruption are very important, andwe cannot ignore the risk of introducing new diseases that can destroy sensitive indigen-ous species. The changes in biodiversity entail a change in relative abundance of species.Moreover, therisktopublichealth should notbeoverlooked asthesespeciesmaybetoxic,such as  ‘ Lagocefalus ’  󿬁 sh, which contains  ‘  Tetrodotoxin ’  a very dangerous substance,capable of causing serious health problems, even death (Demertzis & Iliadis, 2015,2017b). European Union spends at least 12 billion Euros per year, trying to control IASand to overcome their consequences. 1.2. Our research approach Modern IT technologies and innovations are introduced by Computational Intelligence(COIN). These automated high-tech solutions, create the preconditions for designingproper biosecurity and biodiversity protection methods that can evolve and reform theexisting framework. They literally reinforce and simplify IAS detection processes withMachine Learning (ML) identi 󿬁 cation systems. They allow fast and easy data collectionand thus logging of indigenous and non-resident populations in an area. This is achievedby applying an automated process with low  󿬁 nancial requirements. Moreover, they createproper conditions for studying the behaviours of di ff  erent species and their seasonal vari-ation. In other words, they are helping to accurately map the overall invasions. In this way,they are contributing substantially to slowing the uncontrolled expansion of IAS. The sig-ni 󿬁 cance of these intelligent models towards the identi 󿬁 cation and classi 󿬁 cation of IAS hasbeen supported by some recent researches (Demertzis & Iliadis, 2015; Demertzis & Iliadis,2017b, 2017c; Demertzis, Iliadis, & Anezakis, 2017).  This study proposes a new identi 󿬁 cation method for IAS which is presented for the  󿬁 rsttime in the international literature. It is an advanced intelligent Deep Extreme LearningMachine framework (DELM) towards Marine Speci 󿬁 cation Identi 󿬁 cation aided byMachine Hearing (MAHE). The proposed framework uses the Online Sequential Multilayeralgorithm (OSML) plus a Graph Regularized Extreme Learning Machine Autoencoder(GRELMA). It is a highly e ffi cient and srcinal DELE architecture that uses a MultilayerExtreme Learning Machine (MELM) with online learning classi 󿬁 cation capabilities. 2 K. DEMERTZIS ET AL.  Checking whether the recognized item is native or not in its locality, is carried out by aninnovative Geo Location Country Based Services (GCBS) (Demertzis et al., 2017). The implementation of this proposed framework was based on the DELE philosophy.An important aspect of the framework is the usage of ELM which has proven to becapable, of solving a multidimensional and complex IT problem. The Deep ELM simulatesthe functioning of biological brain cells in a most realistic mode. This creates the potentialfor a fully automated con 󿬁 guration of the model with high accuracy classi 󿬁 cation. An inno-vative aspect of this work is the development of this model using Online Sequential Multi-layer algorithm plus a Graph Regularized Extreme Learning Machine Autoencoder thataims to optimize the choice of the input layer weights and bias. This has been done inorder to achieve a higher level of generalization. This method combines two highlye ff  ective machine learning algorithms, for solving a multidimensional and complexmachine hearing problem. Another interesting point is the performance of feature extrac-tion using an intelligent method of audio signal analysis. The proposed method supportsthe autonomous operation of the system comprising the geolocation capability (usingglobal position systems) of the identi 󿬁 ed species and the native ones. This system ispouring arti 󿬁 cial intelligence to digital sensors that can easily (quickly and at low cost)identify invasive or rare species on basis of their phenotypes. This would result in strength-ening biosecurity programmes. 1.3. Literature review   Through a series of new learning architectures and algorithms, have been transformed;DELE methods are now the state-of-the-art in object, speech and audio recognition.Deng and Yu (2013) had proposed methods and applications of DELE. A large deepconvolutional neural network trained (Krizhevsky, Sutskever, & Hinton, 2012) to classify1.2 million high-resolution images in the ImageNet LSVRC-2010 contest into 1000di ff  erent classes. In recent years, Convolutional Neural Networks (CNNs) havebecome very popular and have achieved great success in many computer visiontasks particularly in object recognition. Cellular Simultaneous Recurrent Networks(CSRNs) applied (Alom, Alam, Taha, & Iftekharuddin, 2017) to generate initial  󿬁 ltersof CNNs for features extraction and Regularized Extreme Learning Machines (RELM)for classi 󿬁 cation. Experiments were conducted on three popular data sets for objectrecognition (such as face, pedestrian, and car) to evaluate the performance of the pro-posed system.An object recognition algorithm constructed e ffi cient features automatically withoutrelying on human experts in order to design features for  󿬁 sh species classi 󿬁 cation byZhang, Lee, Zhang, Tippetts, and Lillywhite (2016). Results from experiments showedthat the proposed method obtained an average of 98.9% classi 󿬁 cation accuracy, with astandard deviation of 0.96%. The data set comprised of 8  󿬁 sh species and a total of 1049 images. Moreover, DELE has been the driving force behind large leaps in accuracyand model robustness in audio related domains, like speech recognition.Hinton et al. (2012) presented four successful DELE research approaches for acousticmodelling in sound recognition, namely: Bing-voice-search speech recognition, Switch-board speech recognition, Google voice input speech recognition and YouTube speechrecognition. JOURNAL OF INFORMATION AND TELECOMMUNICATION 3  Moreover (Hinton et al., 2012) proposed the employment of (Deep Neural Networks(DNNs) to extract high-level features from raw data and to show that they are e ff  ectivefor speech and emotion recognition. They  󿬁 rst produced an emotion state probabilitydistribution for each speech segment using DNNs. Then they constructed utterance-level features from segment-level probability distributions. Then these features werefed into an Extreme Learning Machine (ELM), a special simple and e ffi cient single-hidden-layer neural network, to identify utterance-level emotions. The automaticSound Event Classi 󿬁 cation (SEC) has attracted a growing attention in recent years.Feature extraction is a critical factor in SEC system, and DNNs. Actually, they haveachieved the state-of-the-art performance for SEC. The Extreme Learning MachineAuto Encoder (ELMAE) is a new DELE algorithm, which has both an excellent represen-tation performance and it follows a very fast training procedure. A bilinear multi-column ELMAE algorithm has been proposed by Zhang, Yin, Zhang, Shi, and Li(2017) in order to improve the robustness, stability, and feature representation of the srcinal approach. This method was applied towards feature representation of sound signals. Moreover, a similar ELMAE model combined with a two-stage ensemblelearning and classi 󿬁 cation framework was developed to perform the robust ande ff  ective automatic SEC.Additionally, many studies on automatic audio classi 󿬁 cation and segmentation areusing several features and techniques. Zhao et al. (2017) proposed a new method for auto-mated  󿬁 eld recording analysis with an improved segmentation and a robust bird speciesclassi 󿬁 cation. They used a Gaussian Mixture Model (GMM) with an event-energy-basedsifting procedure that selected representative acoustic events. Moreover, they used aclassi 󿬁 cation Support Vector Machine (SVM). 1.4. Paper outline  The remainder of this paper is as follows: A general description of the theoretical back-ground is presented in Section 2. The theoretical concepts of the proposed system are dis-cussed in Section 3. The data sets used and the feature extraction method is included inSection 4. Finally, the marine species identi 󿬁 cation framework and the experimental setupthat was used in this research are discussed in detail in Section 5. Moreover, Section 6 isabout the obtained results which are compared to existing approaches. The conclusionsare summarized and further discussed in Section 7, whereas future research is shown inSection 8. 2. Theoretical background  2.1. Machine hearing Machine Hearing (Lyon, 2017) is a scienti 󿬁 c  󿬁 eld of arti 󿬁 cial intelligence that attempts toreproduce the sense of hearing algorithmically. Audio Signal analysis is the general pro-cedure which refers to the extraction of knowledge based on the correlation amongthe content and the nature of audio signals.In general, the process involves the extraction of certain features capable to di ff  eren-tiate their values, according to the content and the structure of the corresponding 4 K. DEMERTZIS ET AL.
Search
Tags
Related Search
We Need Your Support
Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

Thanks to everyone for your continued support.

No, Thanks