This study presents a deep learning-based method for detecting cracks in concrete structures using the U-Net convolutional neural network. Multiple image datasets, including crack-free noisy images, were used to train and test the model, significantly improving detection accuracy. The findings demonstrate that combining different datasets enhances generalization and real-world applicability. The results suggest that deep learning, especially U-Net, holds potential for autonomous structural damage assessment when integrated with drones and computer vision systems.