Much more information required. A single GS-FSC plot tells us next to nothing. 
Would you share an example micrograph, another with pick locations, 2D classes and the 3D map corresponding to that FSC curve?
Are you sure your helical symmetry parameters are correct? That would be my first guess in a case like this - there are a lot of pitfalls in helical symmetry determination
FSC is bad even before helical refinement, like in non-uniform refinement
Right, but that is not unusual (in my limited experience) for a helical sample. You really often need to figure out the helical symmetry (or generate a very good ab initio reconstruction) before density will become readily interpretable.
The problem for not getting a good ab-initio is probably not enough good particles because sample is highly crowded, I am trying to filter good classes of what i want but it reduces the particle number
Agree with @olibclarke here.
This doesn’t look like a microtubule, but I recently went through the MiCSPARC protocol and going through something similar with whatever your parameters are could help. As someone who does not usually do anything with helical symmetry, I did not realize the amount of sorting and different parameters needed because there can be different helical populations in the same data set with different parameters. As I also learned by going through the protocol, the assignment of the phi angle is important even for two particles that appear to be “on top” of each other, and this can be necessary to get higher resolution structures, even when only using local refinement. It is also possible that your mask is too large if you are trying to refine a single subunit, but I learned almost everything I know in the last couple months from the MiCSPARC folks, so I’ll defer to there or helical experts 
A cryo-EM processing pipeline for microtubules using CryoSPARC
Good luck!