This is a design decision.
You can take the naive approach of using 0.5 as the threshold value, and assign 1 to every value greater than or equal to 0.5, and 0 to the rest.
But you can also, choose your threshold to obtain a classifier that meets certain properties, like achieving a certain false positive rate, o a certain false positive rate. You can see these answers for different objectives when choosing the threshold value: this, this, this.
Note that your question is not specific to the sigmoid activation or to neural networks, but applied also to any probabilistic classifier.