This research utilizes sophisticated Deep Learning (DL) models, namely Inception ResNet, MobileNet, and Xception, to address the task of urban drainage defect detection. A thorough examination is conducted on a single custom dataset to uncover distinct strengths inherent in each model. MobileNet appears as a contender in terms of accuracy, Inception ResNet stands out for its efficiency, and Xception displays robust feature extraction capabilities. Unlike some earlier studies that have focused on single-model performance or specific types of infrastructure, this research provides a comprehensive comparative analysis of multiple DL models tailored specifically for urban drainage systems. While other studies have also explored multiple models, this work uniquely concentrates on the adaptability of these models to drainage defect detection using a single custom dataset. By offering detailed insights into the performance metrics of each model across diverse environmental conditions, this study aids in the informed selection of models for practitioners in the field. The nuanced performance attributes seen in each model contribute to the advancement of urban infrastructure health monitoring, providing a foundation for informed decision-making in the realm of drainage defect detection within urban environments. These findings not only enhance the understanding of DL model capabilities but also contribute to ongoing efforts to monitor and improve the health of urban infrastructure.