Choosing the Right Algorithm: Traditional, Deep Learning, and Transfer Learning

As far as we know, different algorithms have their own advantages and disadvantages, and different algorithms are suitable for different types of problems. It's necessary to choose and optimize according to specific situations. Traditional algorithms are suitable for problems with small data volume and clear features; deep learning is suitable for complex nonlinear problems; transfer learning is suitable for situations where new problems are similar to original problems. They are not isolated, but complement each other. For example, the BSUV-Net deep learning algorithm in literature [2] uses the background determined by traditional methods to perform object detection.