Tourism application with CNN-Based Classification specialized for cultural information

Hirotsu, T.; Hirota, Masaharu; Araki, Tetsuya; Endo, Masaki; Ishikawa, Hiroshi

doi:10.1145/3366030.3366073

Cited by 2 publications

(1 citation statement)

References 22 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…This research is based on these problems and to improve the tourism market's competitiveness. This research proposes an innovative tourism spot identification and recognition mechanism, based on deep learning-based object detection technology, for automatic detection and identification of tourist objects by taking pictures at the location or taking pictures from the internet [4] [5] The efficient solution is to utilize technologies by applying scenery classification based on images. CNN is one of the popular and robust methods to extract various features from images.…”

Section: Introductionmentioning

confidence: 99%

Scenery Classification Using Convolutional Neural Network Towards Indonesia Tourism

Ramadijanti¹,

Karlita²,

Basuki³

et al. 2022

Proceedings of the International Conference on Applied Science and Technology on Social Science 2022 (iCAST-SS 2022)

View full text Add to dashboard Cite

Indonesia is a country that possesses natural wonders and historical buildings, which made Indonesia become one of the popular tourist destination. The scenery classification is a challenging task where the feature distribution from each image may spread. In addition, Indonesian tourism spots are plentiful. In this research, we proposed to utilize deep learning for tourism spot classification using Convolutional Neural Network (CNN) as feature extraction. The dataset consists of different tourism varieties, from man-made objects such as monuments to natural objects such as mountains or beaches. The dataset was selfgathered from the internet and various social media with different angles and does not include any images that dominantly contain people. In addition, in the context of CNN as the basis of feature extractor, we also compared the result with pre-trained CNN architecture trained with Place-365 and ImageNet dataset. The first test was conducted with shallow CNN achieving 48% for the non-augmented dataset and 51% for the augmented dataset. The second test performed with VGG16 and ResNet, combining data augmentation and a pre-trained network. The result reveals data augmentation improves the validation accuracy. Pre-trained VGG16 with Place-365 achieved the highest validation of 90% compared to the other combination. A pretrained network with an augmentation combination improves the model performances significantly by a rough margin.

show abstract

Section: Introductionmentioning

confidence: 99%