I suggest to try first to see what is the difference between the clearly-stained cells and the noisy ones. To reach this goal, you can make a simple pipeline, in which you measure both the intensity and the texture of that specific channel (“Red” in your current pipeline).
You probably need just 1 image that includes both cell types, segment a few good cells for each type (clear vs. noisy) and measure them. I recommend to simply export the result into spreadsheet and use Excel, for instance, to plot the measured values of the 2 populations of cells.
Once you have an idea what is/are the criteria(s) to differentiate between the 2 cell types then we can go back to your main pipeline, build a better filter to exclude the noise cells.
For example, knowing that the noisy cells will have high integrated intensity AND low sum variance of channel Red, you can just use a single FilterObjects module to do this exclusion, no need for ClassifyObjects
Hope that helps.