Built-in Neural Network Algorithms in
GEORADAR-EXPERT
Version 2.2
of the GEORADAR-EXPERT software introduces a new level of automation for ground
penetrating radar data processing through the implementation of built-in neural
network algorithms. The algorithms are integrated into the architecture of the
software and are available to the user within the standard workflow, without
the need to connect additional modules or use third-party software.
The key
advantage of the implemented neural network algorithms is the ability to
perform stable analysis of complex, noisy, and weakly expressed wave patterns
that are typical for real GPR data. In contrast to classical processing methods
based on predefined rules and parameters, neural network algorithms form a
model of the investigated medium by means of neural network structuring of a
large data set represented by a set of wave field attributes obtained during
analysis. This increases the informativeness and reliability of the final
processing result.
The
built-in implementation of neural network algorithms does not require the user
to have specialized knowledge in the field of machine learning. All
computational procedures are performed within the familiar interface of the
software and do not introduce significant changes to the general sequence of
user actions when building a subsurface model presented in the form of an
attribute section. At the same time, the selection of model creation parameters
is simplified and the stability of the result to the influence of various types
of noise is increased.
The
implementation of neural network algorithms in GEORADAR-EXPERT is an important
stage in the development of the software and in the improvement of GPR data
processing methods. The user receives a tool that combines the capabilities of
classical GPR processing with artificial intelligence methods and provides more
accurate, stable, and interpretable results when solving practical tasks.
Below
is an example of GPR profile processing using the built-in neural network
algorithms. The profile was recorded by a GPR unit with a central frequency of
200 MHz on a sea beach. GPR profiling was performed in a direction
perpendicular to the shoreline, at some distance from the water’s edge.
The
investigated ground is characterized by high attenuation of electromagnetic
waves caused by the salinity of coastal deposits. Below, a visual
representation of the neuromodel of the investigated medium is shown in the
form of a system of clusters.
The
neuromodel is formed as a system of clusters representing areas with
statistically homogeneous electrophysical characteristics. Clusters are
identified based on spatial proximity and similarity of properties. As a
result, zones with internal homogeneity of medium parameters are formed.
The
structure of the neuromodel is determined by the number of clusters specified
by the user. The adjustment range includes both a minimum number of clusters
that reflects the main structural elements of the subsurface medium and a more
detailed division into elementary components. Each cluster configuration
defines its own level of generalization of the physical properties of the
investigated object.
The
clusters of the neuromodel are visualized using unique colors. The degree of
similarity between cluster color shades corresponds to the degree of similarity
of medium properties within the respective clusters.
The
final stage of processing of the considered GPR profile is the creation of the
Resistivity attribute section, based on the neuromodel. Below, the result of
exporting this section to a graphical format for subsequent insertion into a
technical report on the GPR survey is shown.
The attribute section provides a clear
representation of the structure of the investigated ground. It shows changes in
the thickness of coastal deposits and the distribution of resistivity in the
direction away from the shoreline. During section creation, corrections based
on the velocity model of the medium formed during the section calculation are
automatically taken into account.
Options for cleaning BSEF analysis data were removed
Due to implementation of neural-network algorithms and modernization of internal software architecture, options for cleaning BSEF analysis results by value repeatability and by deviation from mean values (Mean Deviation Data Cleaning) were removed. This simplified setup of parameters for calculating the subsurface model and for visualizing it as an attribute section.
User
Manual Viewing in GEORADAR-EXPERT Main Window
For quick
access to the user manual of the GEORADAR-EXPERT software, a built-in user
manual browser is implemented in the visualization tab group. Previously, when
selecting the User Manual menu item in the Help menu group, the user manual was
opened outside the software as a compiled help file in CHM format. In this
case, the software window was usually minimized during manual viewing.
In
the new version, when using this menu item, the Help tab opens in the
visualization tab group. The user manual is displayed in a built-in viewer.
This allows the user to obtain reference information for a selected option
while simultaneously viewing the information source and the corresponding
control elements.
