The integrity of concrete structures is crucial for public safety and long-term infrastructure. Nondestructive testing (NDT) methods, like ground-penetrating radar (GPR), are commonly used for inspection but face challenges in analyzing complex radargram data. To improve this, we propose ConcreteNet, a convolutional neural network designed for GPR radargram classification. Based on the "Network in Network" architecture with AlexNet as the base model, ConcreteNet detects structural defects more effectively. It trains on radargram data from the Georgia Southern Engineering Research Building and validates using GPR data from the Georgia Department of Transportation. Additionally, creating a publicly accessible GPR radargram dataset helps researchers train deformity detection models. Benchmarking this dataset against advanced classification and object detection networks further enhances GPR-based deformity detection in concrete.
