Aimed at sorting frequency-hopping networks for under-determined multi-hopping signals, this paper utilizes the temporal sparsity of frequency hopping signals and adopted time-frequency ratio of the original source in hopping signals to estimate hybrid matrix. In calculating the time-frequency matrix, this paper uses the fast overall K average cluster through combination of subspace projection and relative power offset to realize the FH sorting for under-determined condition. Meanwhile, in order to improve the performance of sorting under low SNR condition, the adaptive SNR time-frequency supporting point method is adopted into searching the original single source point. The algorithm is verified by theoretical analysis and simulation.