Recently, the demand for base-isolated structures has been increasing, especially in Japan. For reliable construction of seismic isolation devices, contractors must ensure that the area ratio of the air bubbles occupying the backside of the base plate does not cross the threshold, which is decided by the structural designer. However, the backside of the base plate can include larger and more air bubbles than ordinary concrete surfaces as it is difficult to pour concrete into the foundation properly. Additionally, the inspection process includes many time-consuming tasks. Therefore, it normally takes about one week or longer after concrete work to perform the inspection. We present a method to automate the tasks relating to inspection, including image preprocessing, air bubble extraction, and calculation of the area ratio of the air bubbles, using conventional image processing and a convolutional neural network (CNN) to speed up the inspection. While CNN normally requires a significant amount of training data to achieve high performance, it is generally difficult to collect such a large amount of high-quality data. We have conducted thorough accuracy inspections to evaluate the appropriate amount of training data required. Additionally, we have verified the effect of data augmentation and compared the performance of certain typical CNN architectures. As a result, our method has obtained results that are close to those of manual inspection by a skilled inspector. We can conclude that our method can reduce the overall inspection time by 50% compared to conventional methods.