We systematically investigate the identification and statistical characteristics of foreshock sequences in the North-South seismic belt (21 similar to 43 degrees N, 97 similar to 107 degrees E), which is the test area of China under the "Collaboratory for the Study of Earthquake Predictability (CSEP)," so as to provide data support for foreshock probability forecasting and monitoring deployment. In this paper, we use a stochastic declustering method based on a space-time epidemic-type aftershock sequence (ETAS) model to fit the real catalogs in the North-South seismic belt, and separate each catalog into stochastic versions of family trees, so as to identify the foreshock sequences and analyze their space-time characteristics and statistical characteristics. The results show that the proportion of foreshock sequences of M-L >= 3.0 earthquakes in the North-South seismic belt is not significantly higher than that in other regions of the world (New Zealand, Southern California) without considering the difference in cutoff magnitude. The number of foreshock sequences in the North-South seismic belt displays obvious regional differences, with the number in the middle section distinctly lower and in the south section distinctly higher, while accounting for nearly 50% in all earthquake sequences except for a few areas. The foreshock sequence duration is mostly one to one hundred days, without displaying obvious correlation between the duration of sequences and foreshock sequences and latitude. The background earthquakes have a high b value, approximately 0.89. The b value of foreshock sequences, especially foreshock sequences in the strict sense, is low, approximately 0.72. The b value of aftershock sequences is the lowest, approximately 0.69. The duration of foreshocks in most earthquake sequences is lower than that of aftershocks, especially in sequences with large numbers of events. Meanwhile, sequences that last for more than 1 year have shorter foreshock sequences and longer aftershock sequences.