Pavement crack inspection is crutial for maintaining safe and efficient road infrastructures. Deep Learning particularly Convolutional Neural Network (CNNs) and Vision Transformers(ViTs), hev emerged as powerful tools for automating this process.
● Traditionally used fro crack detection and segmentation. Leverage convolutional layers to extract local features like edges and textures.
● Established models like DeepLabV3+ and U-Net show good performance.
● Limitations: May struggle with long-range dependencies and global context.
● Recent architecture gaining attraction for pavemnt crack inspection. Utilize self attention mechanism to capture long-range dependencies and global context.
● Model like Segformer and UperNet- Swin transformer show promising results.
● Advantages: Potentially better at handling complex and varied crack pattern
● Format: RGB, JPEG ● Resolution: 227 x 227 pixels ● Classes: crack, non-crack ● Total images: 30,000