Identification of Underground Artificial Cavities Based on the Bayesian Convolutional Neural Network

Area of Science:

• Geophysics
• Urban Engineering
• Artificial Intelligence

Background:

• Urban spatial resource exploitation relies on underground artificial cavities.
• Increasing numbers and scales of these cavities necessitate advanced detection methods.
• Geophysical techniques are vital for managing underground structures.

Purpose of the Study:

• To present two methods for identifying underground artificial cavities.
• To improve the efficiency and accuracy of cavity detection.
• To introduce a novel Bayesian convolutional neural network (BCNN) approach.

Main Methods:

• Apparent resistivity imaging using 3D earth models and experimental validation.
• Development of a fast recognition approach based on Bayesian convolutional neural network (BCNN).
• Comparative analysis of BCNN against traditional convolutional neural networks.

Main Results:

• Apparent resistivity imaging effectively identifies underground cavities by comparing simulation data with known cavity positions.
• The BCNN method significantly enhances classification accuracy and efficiency compared to traditional methods.
• BCNN demonstrates superior performance on apparent resistivity image datasets for cavity identification.

Conclusions:

• The developed BCNN offers a highly efficient and accurate solution for underground artificial cavity identification.
• Apparent resistivity imaging combined with BCNN advances geophysical survey capabilities.
• These methods are critical for safe and effective underground engineering and urban planning.