Affected by near-surface absorption, seismic wave energy attenuation and phase distortion lead to a significant reduction in the resolution and signal-to-noise ratio of seismic data. Conventional inverse Q filtering methods have problems such as unstable amplitude compensation and difficult parameter selection. To address these problems, a new unsupervised learning inverse Q filtering high-resolution processing method is proposed. This method integrates the Deep Learning (DL) framework and the seismic wave absorption attenuation theory based on unconditional numerical stability, and provides a DL inverse Q filtering strategy that does not require training labels and avoids numerical instability in amplitude compensation. First, a DL network is constructed, the data to be compensated is input into the network, and the network output is used as the compensated result. Then, the predicted compensation result is sent to the attenuation kernel matrix constructed by the near-surface Q model for forward attenuation. Next, the error between the attenuated seismic data obtained by forward modeling and the original data to be compensated is used to reversely adjust the network parameters, and the error is minimized by iteratively optimizing the network parameters, and the final compensation result is output. In the entire training network prediction process, no data labels are required, and the effect of unsupervised autonomous learning is achieved. The application results of theoretical model data and actual pre-stack seismic data show that compared with the conventional inverse Q filtering method, the unsupervised method can effectively compensate the amplitude energy of seismic signals and has high numerical stability. This method improves the resolution and signal-to-noise ratio of seismic records.