EN | RU    
Scientific & Technical Literature Publishing House
Search:

Home»

Contact Us»

Journals»

News»

Preparation of Papers»

Advertising»

Feedback»

Books»

About Us»



advertisement



Industrial Automatic Control Systems and Controllers

Annotation
<< Back
Developing a Hibrid Neural Network with Local Neural Units for Collaborative Filtering Tasks
E.B. Cozac

The article presents the results of research and analysis of expert information based on computer information processing methods for recognizing graphic objects in various conditions using deep neural networks. The present work presents a hybrid deep neural network method integrating a global multilayer deep neural network with local blocks (neural networks). The focus is on two neural architectures, ResNet50 and VGG19. This study presents the results of analyzing the operation of deep neural network models for performing information-theoretic analysis and filtering arrays of graphical information. Experiments have been carried out to test the performance and efficiency of the developed neural network. The aim of the work is to use deep neural networks for image classification. Image recognition systems have been trained and tested on a representative database of 10.000 people. We used photographs of dogs and cats for the experiment, which greatly simplified the analysis due to the ease of their interpretation by humans. The paper also investigates how the appearance of atypical input signal (photos) affects the output signal of the neural network classification mechanism. Misclassified photographs were analyzed and matched against interpretations made by real people. System evaluation was based on statistical metrics, such as precision, sensitivity, specificity, and ROC curves. The accuracy of all neural networks was also measured. The highest accuracy was achieved with ResNet50 and VGG19 neural networks. Recognition results were very high with AUC ROC ≥ 0,8 and 90 % accuracy. Moreover, the most accurate recognition results were those with data expansion.
Keywords: artificial neural networks; hybrid neural networks; fuzzy logic; image recognition.


DOI: 10.25791/asu.9.2021.1309

Pp. 19-29.

 Sections

«About journal

«Archive

«Thematic focus of the journal

«Formatting rules

«Stages of the review and publication

«Review process

«Editorial and Professional Ethics

«Detecting plagiarism

«Editors and Editorial Board

«News journal


 Journals
...................................
Instruments and Systems: Monitoring, Control, and Diagnostics
...................................
Instrument-Making and Automation Means. Encyclopedic Textbook
...................................
Industrial Automatic Control Systems and Controllers
...................................
Ecological Systems and Devices
...................................
Aerospace Instrument-Making
...................................
Engineering Physics
...................................
History of Science and Engineering
...................................
Music and Time
...................................
Note Album
...................................
Musicology
...................................
Universal History
...................................
Directory of engineer
...................................
Applied Physics and Mathematics
...................................
News Academy of Engineering Sciences A.M. Prokhorov
...................................

Last news:

Выставки по автоматизации и электронике «ПТА-Урал 2018» и «Электроника-Урал 2018» состоятся в Екатеринбурге

Открыта электронная регистрация на выставку Дефектоскопия / NDT St. Petersburg

Открыта регистрация на 9-ю Международную научно-практическую конференцию «Строительство и ремонт скважин — 2018»

ExpoElectronica и ElectronTechExpo 2018: рост площади экспозиции на 19% и новые формы контент-программы

Тематика и состав экспозиции РЭП на выставке "ChipEXPO - 2018"

   Rambler's Top100 Rambler's Top100         


    Management system developed by: ananskikh.ru
© Publishing House "NAUCHTEKHLITIZDAT", 2005-2024