This study systematically investigates the influence of initial stress states and reloading time intervals on the stress memory capacity of granite. Uniaxial compression acoustic emission (AE) tests were conducted, in which AE signals exhibiting the Kaiser effect were effectively identified using a cluster analysis-based signal processing method, with the reliability of the classification further verified by a random forest algorithm. A time–frequency domain "dual-energy" criterion was established to determine the Kaiser point by identifying mutation points in both energy and cumulative ring-down counts in the time domain, as well as transition points in energy spectral density in the frequency domain. This approach significantly enhances the noise resistance and robustness of stress memory identification. Experimental results indicate that granite retains a strong stress memory when the initial stress is 30% or 40% of the peak stress and the reloading interval does not exceed 14 days. However, this capacity gradually diminishes when the interval exceeds 30 days. Under an initial stress of 50% of the peak stress, no effective Kaiser effect is observed, regardless of the reloading interval. This research provides a more reliable analytical method for in-situ stress measurement in deep rock masses.