How to avoid being (falsely) accused of plagiarism when reusing notations and definitions

It is generally known that definitions and restated theorems can cause significant textual overlap between papers to an extent that could be considered problematic. There is no point in worrying about automatic plagiarism checks for journal submissions: If a journal is relying on that alone without an editor checking, it is not a reasonable journal to submit to anyway.

Don't copy that part of the other paper word-for-word. Just introduce the notations and definitions in your own words.

It becomes plagiarism only if you intentionally present [parts of] someone else's work as your own, otherwise it is citation. Robots may not be smart enough to distinguish, but that's why nobody relies on them much: they help to make decisions, not make decisions by themselves. What those automatic systems do is mark parts of your work as resembling parts of some other works to attract attention of human reviewers, who decide if that is plagiarism or not. Don't bother rephrasing the quotes just to game those systems.

I saw one of such systems at work, it would paint shorter indirect quotes yellow, indicating it has "79% overlap with [7]", a long direct quote it would paint red, because it's "100% overlap" with [19]", then add something like another "References" section, where those [7] and [19] would be, so the reviewers can examine them, if they want. Would be no problem for you if you made it clear that those are quotes and provided the references. (Also, it painted the entire "Reference" section red, because every reference was already used in like 200 other papers, but of course nobody takes that as plagiarism.)

The reviewers may still reject the paper because it has "too much quotations/too little original content", but that's a different story.