 We developed a system based on texture analysis, which worked very well. We have a working software system that is being tested by some nuclear power plants. Once we did that, and the results were very impressive, we moved to the next level, which we used deep learning. Instead of processing just a single image, we process all the frames in a video. A crack might not be detected in some frames, but detected in some other frames. So how we aggregate the information that we extract from these frames is another challenge. And we use Bayesian decision theory to enhance the performance of our crack detection algorithm.