🛣 Pavement Crack Inspection using CNNs & Vision Transformers

🔍 Asphalt Crack Detection with Deep Learning (CNN & ViT)

---

📌 Description

Inspecting and identifying asphalt surface cracks is essential for enhancing road safety.
This project leverages deep learning to automatically detect road cracks.
✅ Advantages:

• Reduced cost and time for road inspections
• Higher accuracy compared to traditional methods
• Capability for large-scale analysis

---

📌 Methods Used

1️⃣ Convolutional Neural Networks (CNNs)

✔️ Traditional models in computer vision
✔️ Extraction of local features such as edges and crack textures
✔️ Tested models: U-Net, DeepLabV3+

2️⃣ Vision Transformers (ViTs)

✔️ Latest technique in image processing
✔️ Utilization of Self-Attention for better crack understanding
✔️ Tested models: Segformer, UperNet-Swin Transformer

---

📦 Installation and Setup

Dependencies

Libraries used in this project:

• PIL (Pillow)
• collections.Counter
• keras.preprocessing.image
• scikit-learn (sklearn)
• tensorflow.keras
• cv2 (OpenCV)
• matplotlib
• numpy
• pandas
• plotly.express
• seaborn
• tensorflow
• tensorflow_addons
• visualkeras
• warnings

Run the Notebook

jupyter notebook sentiment-analysis-bert-xai.ipynb

---

📊 Sample Results

Model	Accuracy	Advantages	Disadvantages
U-Net	85%	High accuracy for clear cracks	Requires more data
DeepLabV3+	88%	Better performance on irregular edges	Slower speed
Segformer	91%	Precise detection of crack textures	Heavier computations

---

• If you have suggestions for improving the project, please submit a Pull Request.
• To report issues, please open an Issue.

---