No, you can't avoid the conversion. featureCounts is trying to protect the names from systems that can't handle punctuation in variable names, but I agree it is not necessary here.

Of course you had to input the file names to featureCounts in the first place:

fc <- featureCounts(files, ...)

so you can easily put them back at the end:

colnames(fc$counts) <- files

I personally like to use

colnames(fc$counts) <- limma::removeExt(basename(files))

Why don't you simply grep the sample names from disk and then put them as colnames back into the output? Something like this:

tmp.files <- list.files(path = "~/Desktop/", pattern = ".bam", full.names = TRUE)

tmp.names <- sapply(strsplit(tmp.files, split="\\/"), function(x) rev(x)[1])

countmatrix<-featureCounts(files = tmp.files,
                           annot.ext = "~/Desktop/test.saf")

colnames(countmatrix$counts) <- c(tmp.names)
colnames(countmatrix$stat)   <- c("Status", tmp.names)