Quantitative Evaluation of Plant and Modern Urban Landscape Spatial Scale Based on Multiscale Convolutional Neural...

Area of Science:

• Environmental Science
• Computer Science
• Urban Planning

Background:

• Urban landscapes are critical ecosystems for city sustainability.
• Effective landscape information extraction is vital for urban planning and development.
• Existing methods may lack the precision needed for detailed urban ecosystem analysis.

Purpose of the Study:

• To propose and evaluate a novel landscape information extraction model using deep convolutional neural networks (CNNs).
• To investigate the efficacy of multiscale CNNs for classifying urban landscape features.
• To quantitatively assess the performance of the developed deep CNN model for urban landscape analysis.

Main Methods:

• Development of a landscape information extraction model based on deep convolutional neural networks.
• Implementation of a multiscale CNN approach for landscape classification.
• Quantitative evaluation using confusion matrix, production accuracy, user accuracy, and kappa coefficient.

Main Results:

• The proposed multiscale CNN model achieved an overall kappa coefficient of 0.91 and 93% classification accuracy.
• Water target identification exceeded 90%, with user accuracy at 99.78% and production accuracy at 91.94%.
• The model demonstrated superior performance compared to other methods, with the best overall accuracy and kappa coefficient.

Conclusions:

• The deep convolutional neural network model provides a highly accurate method for urban landscape information extraction.
• The multiscale CNN approach is effective for classifying complex urban landscape features.
• This study offers valuable insights for the quantitative evaluation of urban landscape spatial scales and supports sustainable urban development.