Using deep learning to quantify the beauty of outdoor places

Area of Science:

• Environmental Psychology
• Computer Vision
• Urban Planning

Background:

• Protected natural landscapes are linked to improved public health.
• Understanding the aesthetic qualities of outdoor environments is crucial for well-being and policy.
• Distinguishing between naturalness and beauty in outdoor spaces requires investigation.

Purpose of the Study:

• To identify key features contributing to the perceived beauty of outdoor environments.
• To explore the role of both natural and man-made elements in scenicness.
• To develop a computational model for identifying scenic locations.

Main Methods:

• Analysis of over 200,000 user ratings from the 'Scenic-Or-Not' online game.
• Feature extraction from images using the Places Convolutional Neural Network (CNN).
• Training a neural network to predict scenicness based on image features.

Main Results:

• Scenic outdoor locations are characterized by natural features (e.g., coast, mountains) and man-made structures (e.g., towers, castles, viaducts).
• While trees enhance scenic ratings, generic green elements like grass and athletic fields do not.
• A trained neural network effectively identifies scenic places, recognizing both natural and built environments.

Conclusions:

• Online data and AI provide valuable insights into human aesthetic preferences for outdoor environments.
• Both natural landscapes and specific built structures contribute to perceived environmental beauty.
• Findings offer quantitative data for policymakers in environmental design and conservation efforts.