This thesis explores the use of deep learning, specifically Convolutional Neural Networks (CNNs), for automated pavement crack segmentation in North Cyprus, addressing the need for efficient road maintenance. The study emphasizes the limitations of manual inspection and introduces a CNN-based U-Net architecture, developed to automate feature extraction, and enable more accurate and efficient crack segmentation. Implementing a U-Net model with custom layers to learn features without transfer learning forms the core of this methodology, utilizing ReLU and sigmoid activation functions, binary cross-entropy as the loss function, and the Adam optimizer. The model is evaluated using metrics including accuracy, precision, recall, F1 score, and IoU, achieving over 98% accuracy.
