Deep Learning for Billboard Classification

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

8 Downloads (Pure)


Advertising is essential to increase product awareness and foster a positive outlook, which in turn helps sales. To promote the brand and its products, billboard advertisements are widely used. This paper presents a novel approach for classifying billboards. The proposed method utilises Convolutional Neural
Network (CNN) architectures to extract features from the images to enable classification. The model is trained on a dataset of billboards collected from various locations and achieves results that demonstrate high classification accuracy. The system is trained and evaluated using the CIFAR10 dataset, which includes 10 classes of objects and an additional 11th class - ’billboard’,
is included. The experiment utilises five different CNN architectures: Basic CNN, ResNet, VGG, MobileNet, and DenseNet. The performance and evaluation of each architecture are presented in detail, and extensive experiments and comparisons are conducted to determine the most effective model for classifying billboards. The results indicate that a CNN and its architectural designs are a promising solution for automating the classification of billboards in the wild.
Original languageEnglish
Title of host publicationDIGITAL 2023
Subtitle of host publicationAdvances on Societal Digital Transformation
EditorsPetre Dini
PublisherInternational Academy, Research, and Industry Association
Number of pages7
ISBN (Print)978-1-68558-115-2
Publication statusPublished online - 11 Oct 2023
EventAdvances on Societal Digital Transformation DIGITAL 2023 - Portugal, Porto, Portugal
Duration: 25 Sept 202329 Sept 2023


ConferenceAdvances on Societal Digital Transformation DIGITAL 2023
Abbreviated titleDIGITAL 2023
Internet address


  • Classification
  • CNN-Architecture
  • Billboard
  • Image-Processing
  • CIFAR10


Dive into the research topics of 'Deep Learning for Billboard Classification'. Together they form a unique fingerprint.

Cite this