Active Actions in the Extraction of Urban Objects for Information Quality and Knowledge Recommendation with Machine...

Area of Science:

• Geoinformatics and Remote Sensing
• Urban Planning and Sustainability
• Machine Learning for Geospatial Analysis

Background:

• Increasing urban development necessitates smart city technologies for effective land monitoring and ecological process understanding.
• A critical gap exists in technologies that validate information quality for strategic urban planning.
• Managing large geospatial datasets for quality information access demands significant effort.

Purpose of the Study:

• To develop and evaluate methods for mapping land use zones (vegetation, soil, asphalt, buildings) in Itajaí, Brazil.
• To address the need for reliable land monitoring tools supporting sustainable urban development.
• To provide a technological infrastructure for informed conservation and management decisions.

Main Methods:

• Object-based image analysis using classifiers (OneR, NaiveBayes, J48, IBk, Hoeffding Tree) with GeoDMA.
• Implementation of deep learning models: Region-based Convolutional Neural Network (R-CNN) and YOLO algorithm.
• Data processing and cataloging for geospatial object identification and similarity analysis.

Main Results:

• Achieved a classification accuracy of 85% and a Kappa agreement coefficient of 76% for land zone identification.
• Successfully characterized vegetation, exposed soil, asphalt, and buildings in urban and rural areas.
• Demonstrated the effectiveness of integrating machine learning classifiers and deep learning for geospatial object recognition.

Conclusions:

• The study provides a practical approach to extracting valuable information from geospatial data for urban management.
• The developed methods enhance the accuracy of land use mapping, supporting conservation and management decisions.
• A foundational technological infrastructure was established to aid sustainable urban development and decision-making.