Most experiments detect objects. In theory, you can have your own dataset to detect patterns but this is not a beginner task. If you don’t have enough time or expertise resource, the chance of failure is very high. But if you manage to pull it off, that will be an interesting application in DL (very interesting). You first task involving collecting dataset, likely >10K or >100K samples. Then you train the model which you will likely encounter issues not found in object detection. But I think you should try non deep learning image segmentation first. It sounds like you want out-of-the-box technology.