At present, the detection data set in my hand has background in addition to the target. If classification is done, there should only be the target in the data set. So I wonder, do I still need to scratch out the target for the classification data set?