Land use classification of open-pit mine based on multi-scale segmentation and random forest model

Area of Science:

• Environmental Science
• Remote Sensing
• Geospatial Analysis

Background:

• Mining production is a key Chinese industry but causes significant ecological issues.
• Earth observation technology offers solutions for monitoring mining-induced land changes.
• Conventional remote sensing methods struggle with accuracy due to image noise.

Purpose of the Study:

• To develop and evaluate an advanced method for extracting land occupation information from open-pit mining areas.
• To improve the accuracy of monitoring environmental impacts associated with mining activities.
• To compare the effectiveness of a novel approach against conventional methods.

Main Methods:

• Utilized a multi-scale segmentation algorithm with a layered mask approach for object-based image analysis.
• Employed the random forest model, incorporating spectral, geometric, and texture features for classification.
• Selected 23 characteristic factors across spectral, geometric, and texture domains.

Main Results:

• The proposed method achieved an overall extraction accuracy of 86% and a Kappa coefficient of 0.78.
• This represents a significant improvement over the conventional method's accuracy of 79% and Kappa of 0.68.
• Successfully extracted detailed land occupation information related to mining production.

Conclusions:

• The object-based approach combined with the random forest model offers superior accuracy for monitoring mining land occupation.
• This advanced technique effectively overcomes limitations of traditional pixel-based methods.
• The findings provide a more reliable tool for environmental management in mining regions.