As a new type of remote sensing detection technology, hyper-temporal photometric detection is characterized by high accuracy and timeliness. The collected hyper-temporal photometric signals contain characteristic information such as the working status and structure of the target. Aiming at the non-stationary and non-linear characteristics of hyper-temporal signals, this paper adopts Variational Mode Decomposition (VMD) for signal processing, and correlation indicators are used to filter mode components to achieve noise reduction. Aiming at the problem that the algorithm performance is seriously affected by key parameters such as the number of modes and the quadratic penalty parameter when processing non-stationary signals, an adaptive variational mode decomposition (AVMD) algorithm based on energy difference is proposed. The mutation point is identified by the second derivative of the energy difference curve to determine the number of decomposition modes, and the quadratic penalty parameter is further determined based on the principle of energy difference minimization. Verified through simulation data and measured vibration photometric signals of momentum wheels, the proposed algorithm achieves precise time-frequency feature extraction and effectively suppresses mode aliasing. The denoising results of simulated signals indicate that the Signal-to-Noise Ratio (SNR) is improved by 40.43%, while the root mean square error (RMSE) is reduced by 55.08%, providing a reliable signal processing approach for hyper-temporal photometric detection.