Internal Pipe Corrosion Assessment with Ultrasound and CNNs

This study introduces a dual-mode methodology for quantifying pipe corrosion by employing ultrasound technology in conjunction with convolutional neural networks (CNN). The codes include components for data handling, model development, preprocessing, and utility functions.

---

Components

• Data:

  • data/: Directory containing the original ultrasound data.

• Data Handling:

  • data.py: Functions for loading datasets.

• Model Development:

  • model/: Directory containing various CNN architectures:
    • AlexNet.py
    • DenseNet.py
    • EfficientNet.py
    • InceptionNet.py
    • ResNet.py
    • VGG.py
    • __pycache__/: Cached Python bytecode files.

• Preprocessing:

  • preprocess.py: Implementation code for data preprocessing.
  • preprocess_fn.py: Functions required for preprocessing.

• Execution Scripts:

  • run_experiment.py: Main execution script.
  • training.py: Contains classes and functions related to training.
  • utils.py: Collection of utility functions.

• Output and Results:

  • confusion_matrix.zip, output.hwp, plot.zip, vgg_loss_plot.zip: Output files and archives related to results and plots.
  • draw_acc-lossplot.py: Script for plotting accuracy and loss.

---

Usage

• Run run_experiment.py to execute the main functionalities of the project.
• The model/ directory contains implementations of different CNN architectures.

---

Experimental Results

Validation Accuracy and Loss

Confusion Matrix

effcientNet

• 1L(3class)
• 50ml(3class)
• 6class

InceptionNet

• 1L(3class)
• 50ml(3class)
• 6class

VGGNet

• 1L(3class)
• 50ml(3class)
• 6class

ResNet

• 1L(3class)
• 50ml(3class)
• 6class

---

---