At the population level, astrocytes dynamically interact with neurons.^{1 – 3} Astrocytes sense activity within the neuronal network and respond to neural signals by generating ${\mathrm{Ca}}^{2+}$ activities in vivo.^4,5 Simultaneously, astrocyte activation or manipulation modulates neuronal network activity.^2,3 This bidirectional, interconnected communication between astrocytes and neurons is collectively termed the “astrocyte-neuronal network.”^6,7 Accumulating experimental evidence suggests that various behaviors result from coordinated activity between neurons and astrocytes within these networks.^2,8,9 Therefore, deciphering the interaction between astrocytes and neurons within astrocyte-neuronal networks is essential for comprehensively understanding the genesis and dysregulation of behavior.^2,3,10

Two-photon microscopy enables in vivo imaging of astrocytic Ca²⁺ activity, yet detecting weak, transient, and background-coupled signals remains challenging due to low signal-to-noise ratios and heterogeneous noise. Here, we propose a low-parameter, adaptive framework for detecting weak astrocytic Ca²⁺ signals in two-photon imaging. After short-window frame accumulation, static background suppression, and Gaussian smoothing to stabilize intensity statistics, signal candidates are identified via segment-wise Gaussian mixture modeling, temporal persistence masking, and adaptive threshold updates. In simulated videos, the proposed method improved the Dice coefficient from 0.06 to 0.77 and increased the reference SNR from −9.82 to 3.40 dB. In in vivo recordings, the local SNR increased from 5.58 to 7.28 dB. Compared with fixed thresholding, AQuA, and AQuA2, our method was more robust under high-noise conditions while requiring only three user-defined parameters (minimum area, minimum duration, and an initialization coefficient). This framework provides an interpretable and computationally practical front-end module for the robust extraction of astrocytic Ca²⁺ signal in low-SNR two-photon imaging.