GPR Viewer Application (v00readerV3.py)

@Mrinal Kanti Layek @layek.mk@gmail.com

Summary of Fixes and Enhancements

This document summarizes the fixes and enhancements made to the GPR viewer application.

---

1. SciPy/PyHHT Compatibility Fix

Problem

ImportError: cannot import name 'angle' from 'scipy'

The pyhht library (v0.1.0) attempts to import scipy.angle, which was removed in SciPy 1.12+ and moved to numpy.angle.

Solution

Added a compatibility patch at the top of the file (before importing pyhht):

import scipy
if not hasattr(scipy, 'angle'):
    scipy.angle = np.angle

from pyhht.emd import EMD

Location

Lines 14-23 in v00readerV2.py

---

2. Decluttering Modules (MS, SVD, RNMF)

Description

Added three GPR clutter removal methods based on the paper:

Ge et al. 2024, IEEE TGRS - "Wavelet-GAN: A GPR Noise and Clutter Removal Method Based on Small Real Datasets"

New GUI Buttons

Three new buttons added to the interface (column 15):

• MS - Mean Subtraction
• SVD - Singular Value Decomposition
• RNMF - Robust Non-negative Matrix Factorization

Method Details

2.1 Mean Subtraction (MS)

• Function: ms_declutter_dialog(), _apply_mean_subtraction()
• Algorithm: Subtracts the mean trace from all traces
• Use Case: Removes horizontal clutter (direct wave, ringing noise with strict horizontal distribution)
• Parameters: None (one-click operation)

2.2 SVD Declutter

• Function: svd_declutter_dialog(), _apply_svd_declutter()
• Algorithm: Uses Singular Value Decomposition to separate clutter from signal
• Use Case: Removes first N singular value components which typically capture horizontal clutter
• Parameters:
  • Components to remove (default: 1, range: 1-20)

2.3 RNMF Declutter

• Function: rnmf_declutter_dialog(), _apply_rnmf_declutter()
• Algorithm: Robust Non-negative Matrix Factorization
• Use Case: Separates low-rank clutter from sparse target responses
• Parameters:
  • Number of components/rank (default: 5)
  • Max iterations (default: 200)
  • Sparsity weight alpha (default: 0.1)
• Note: Includes progress bar during processing

Key Features

• Original raw data preserved (all processing on copies)
• Each method adds process tag to history (MS, SVD-N, RNMF-N)
• Reset button restores original data
• Full backward compatibility with existing codebase

---

3. Predictive Deconvolution Fix

Problem

The original deconvolution was removing almost all the signal:

• The prediction filter was predicting the entire signal (not just ringing)
• Subtracting prediction left near-zero residuals
• Only the top portion of data was visible; everything below was flat gray

Solution

Complete rewrite of predictive_deconvolution() with proper RADAN-style implementation:

Key Changes

1. Proper Prediction-Error Filter (PEF):

   # PEF = [1, 0, ..., 0, -f0, -f1, ..., -fn]
   pef = np.zeros(lag + n)
   pef[0] = 1.0  # Unit spike preserves signal
   pef[lag:lag + n] = -f  # Negative filter removes predictable ringing

2. Amplitude Preservation:

   # Normalize output to match input amplitude
   filtered = filtered * (trace_std / filtered_std)

3. Better Autocorrelation:

   • Uses unbiased autocorrelation estimate (divides by overlap count)
   • Better numerical stability

4. Improved Regularization:

   • Adds prewhitening to both autocorr[0] and the diagonal of R matrix
   • Prevents singular matrix issues

Updated Default Parameters

Parameter	Old Value	New Value
Prediction lag	8	4
Prewhitening	0.08	0.1
Overall gain	4.0	1.0
Start sample	60	0

Location

• Function: predictive_deconvolution() in GPR2DLoader class
• Dialog: deconvolution_dialog() in GPR2DViewer class

---

4. Wiggle Plot Enhancement (RADAN7-Style)

Problem

Basic wiggle implementation lacked features found in commercial software like RADAN7.

Solution

Complete rewrite with professional-grade features:

New Features

1. Parameter Dialog (plot_wiggle()):

   • Interactive settings before plotting
   • Real-time parameter adjustment

2. Trace Decimation:

   • trace_skip parameter (1=all, 2=every 2nd, etc.)
   • Auto-adjusts for large datasets (default: n_traces // 200)

3. Variable Area Fill Modes:

   • positive - Standard VA (fill positive peaks)
   • negative - Fill negative peaks
   • both - Fill both with different colors
   • none - Wiggle lines only

4. Color Options:

   • Positive fill color (black, blue, red, gray, darkblue)
   • Negative fill color (white, red, blue, gray, lightgray)
   • Background color (white, lightgray, beige, black)

5. Display Controls:

   • Amplitude scale factor
   • Clip percentage (clips outliers)
   • Line width control
   • Show/hide wiggle lines
   • Fill opacity (0-1)

6. Professional Rendering:

   • 99th percentile normalization (avoids outlier distortion)
   • Proper trace spacing calculation
   • Depth grid lines
   • Automatic axis scaling

Parameters

Parameter	Default	Description
Trace skip	auto	Plot every Nth trace
Amplitude scale	1.0	Scale factor for amplitudes
Clip (%)	100	Clip at % of max amplitude
Fill mode	positive	Which peaks to fill
Positive color	black	Fill color for positive
Negative color	white	Fill color for negative
Line width	0.5	Wiggle line thickness
Show lines	True	Display wiggle traces
Fill opacity	0.85	Fill transparency
Background	white	Plot background color

Location

• Dialog: plot_wiggle() method
• Renderer: _render_wiggle() method

---

Dependencies

Required Python packages:

numpy
scipy
matplotlib
tkinter
pandas
pyproj
geopandas
shapely
contextily
scikit-learn
pyhht
pywt
opencv-python (cv2)
scikit-image
folium (optional)

Installation

pip install numpy scipy matplotlib pandas pyproj geopandas shapely contextily scikit-learn pyhht pywt opencv-python scikit-image folium

---

Usage

Run the application:

python v00readerV2.py

Workflow

1. Load Data: Use file browser to load .v00, .d00, or .dt files
2. Apply Processing: Use toolbar buttons for various processing options
3. Declutter: Use MS, SVD, or RNMF buttons for clutter removal
4. Deconvolution: Click "Deconvolution" button to remove ringing
5. Visualization: Switch between image, wiggle, and other plot modes
6. Reset: Click "Reset" to restore original data

---

File Structure

v00readerV2.py
├── GPR2DLoader class
│   ├── Data loading (v00, d00, dt, HDR, GEOX)
│   ├── predictive_deconvolution()  [FIXED]
│   ├── Processing methods
│   └── Data management
│
└── GPR2DViewer class
    ├── GUI setup
    ├── Plot functions
    │   ├── plot_gpr() - Image display
    │   ├── plot_wiggle() - Wiggle dialog [ENHANCED]
    │   └── _render_wiggle() - Wiggle renderer [NEW]
    ├── Decluttering modules [NEW]
    │   ├── ms_declutter_dialog()
    │   ├── _apply_mean_subtraction()
    │   ├── svd_declutter_dialog()
    │   ├── _apply_svd_declutter()
    │   ├── rnmf_declutter_dialog()
    │   └── _apply_rnmf_declutter()
    └── deconvolution_dialog() [FIXED defaults]

---

Version History

Date	Changes
2026-01-29	Added MS, SVD, RNMF decluttering modules
2026-01-29	Fixed pyhht/scipy compatibility issue
2026-01-29	Fixed predictive deconvolution (amplitude preservation)
2026-01-29	Enhanced wiggle plot (RADAN7-style features)

---

References

1. Ge et al. 2024, IEEE TGRS - "Wavelet-GAN: A GPR Noise and Clutter Removal Method Based on Small Real Datasets"
2. GSSI RADAN 7 User Manual
3. Yilmaz, O. (2001). Seismic Data Analysis (for deconvolution theory)