regions of high similarity between genomes are not called duplicated sequences

those are just that - similar regions

now in your case don't worry about this at all, your mapper will flag the multiply mapped reads in some way (for example bwa will set the mapping quality to zero and will produce an XA tag I think) thus from the alignments you can always tell what is the coverage of the unique regions, just filter for mapping quality to use the uniquely mapped reads

This is precisely why I added this comment to your last post: C: how can i determine the couverture (coverage ) of my reads on the whole genome u

There is no way for you to remove "duplicates" (similar sequences) in reference genomes (unless you have multiple strains of one organism and can cut those down some).

Since your aim seems to be calculation of "coverage" with an aligner like BBMap you can handle the multi-mapping reads in several ways. Following is the relevant option.

ambiguous=best          (ambig) Set behavior on ambiguously-mapped reads (with 
                        multiple top-scoring mapping locations).
                            best    (use the first best site)
                            toss    (consider unmapped)
                            random  (select one top-scoring site randomly)
                            all     (retain all top-scoring sites)

You could choose random to conservatively place them at one random location. This may be the best (not what best means above) option. If you choose to place them in all locations then you would be seriously over-estimating your coverage (since the read likely came from one genome). Other two options would just result in odd placements or loss of a lot of data.

Note: I don't think bowtie2 has a similar set of options. It may simply stop aligning reads to multi-mapping locations after a certain number is reached.