Early detection of fissuring in concrete structures has become an increasing necessity in civil engineering, as it directly contributes to ensuring structural safety, durability, and the adequate maintenance of infrastructure. Nevertheless, conventional methods based on visual inspections exhibit significant limitations related to high costs, subjectivity, and low operational efficiency. Within this context, the present research aims to develop and evaluate the performance of two convolutional neural networks (CNNs) for fissure detection in concrete structures. This study was conducted under a quantitative, applied, and experimental research approach. A database consisting of 8000 field-acquired images was constructed; these images were subsequently preprocessed and employed to train and validate the VGG16 and ResNet50 models using deep learning techniques. Model performance was assessed using accuracy, precision, and sensitivity metrics and further validated in a real-world scenario corresponding to a reinforced concrete bridge located in Manta. The results demonstrate that both models achieve satisfactory performance in the automated evaluation and detection of surface fissures. ResNet50 outperformed VGG16 in terms of overall accuracy and sensitivity, whereas VGG16 exhibited higher precision in classification tasks. These findings confirm the potential of deep learning as an efficient and objective tool for structural health monitoring, with direct implications for optimizing inspection procedures, control strategies, and preventive maintenance processes.