Detecting cracks in concrete is a critical aspect of structural health monitoring, directly contributing to the goals of sustainable development by enhancing infrastructure safety, reducing resource-intensive repairs, and extending the lifespan of built environments. This study investigates the potential of machine learning techniques—specifically advanced convolutional neural networks (CNNs)—to accurately and efficiently identify concrete cracks. A robust preprocessing phase is employed to improve crack visibility. Two well-established CNN architectures, ResNet-50 and VGG-16, are integrated to boost detection performance. ResNet-50 excels at identifying complex crack patterns, while VGG-16 offers reliable accuracy with lower computational requirements for standard-resolution images. While each model is effective on its own, their combined use significantly improves detection accuracy. The proposed hybrid model outperforms individual implementations of ResNet-50 and VGG-16, supporting the development of safer, more durable, and resource-efficient infrastructure systems aligned with sustainable development objectives.