Deep learning based real-time tourist spots detection and recognition mechanism.

Area of Science:

• Computer Science
• Artificial Intelligence
• Tourism Technology

Background:

• Increasing reliance on visual information for travel decisions.
• Challenges for tourists in identifying attractions from images and accessing relevant information.
• Need for advanced systems to bridge the gap between visual recognition and textual data for tourism.

Purpose of the Study:

• To develop an innovative tourist spot identification mechanism using deep learning.
• To enable real-time detection and recognition of tourist attractions from images.
• To enhance the tourism market's competitiveness through intelligent information access.

Main Methods:

• Implementation of a tourist spot recognition system utilizing the You Only Look Once version 3 (YOLOv3) object detection model.
• Development on the Tensorflow AI framework for image-based identification.
• Testing and validation using a dataset of tourist spots in Hsinchu City, Taiwan.

Main Results:

• The YOLOv3 system successfully identified 28 tourist spots in Hsinchu.
• Achieved a recognition time of 4.5 seconds and a mean average precision of 88.63% at IoU=0.6.
• Demonstrated superior efficiency and precision compared to Faster region-convolutional neural networks (Faster R-CNN) and Single-Shot Multibox Detector (SSD) models.

Conclusions:

• The proposed deep learning-based tourist spot identification system is effective and efficient.
• YOLOv3 offers a significant performance advantage over Faster R-CNN and SSD for tourist attraction recognition.
• The system facilitates enhanced travel planning by integrating with open data platforms for itinerary generation and navigation.