Hi,
Yes! Take a look at this reply and adapt the parameters to your problem.
However, assuming that you are going to filter sample-wise, be aware that to filter all taxa with less than 3% relative abundance in any sample may be a little too stringent (you may end up filtering all taxa). Other concern is that it would greatly distort the relative abundance of other taxa, especially for high diversity samples, besides losing a lot of data.
Maybe the best choice would be to export the barplot raw data and replot assigning "Other" for the taxa that are below your threshold. This way you don't disturb the relative frequency of other taxa and get rid of endless legends.
Good luck,
Vitor