Wavefront sensor-less adaptive optics (WFS-less AO) has emerged as a pivotal area of research in recent years. The dynamic targets and random disturbances introduced by atmospheric turbulence pose significant challenges for traditional image-based wavefront sensing techniques. Current methods typically require multiple images to decouple targets and atmospheric turbulence features. In contrast, this paper proposes a novel target-independent wavefront sensing method that utilizes a single distorted image, incorporating frequency feature extraction and the attention mechanism. The proposed method employs a frequency network that separates target and turbulence information into distinct frequency bands, allowing for specialized processing of each component. Additionally, a multi-head attention mechanism is used to extract local features and capture global dependencies, improving the modeling of complex distortions. To evaluate the proposed method, a simulated dataset of degraded images is created for training and testing. Experimental results demonstrate that the root mean square error (RMSE) between the predicted and true wavefronts ranges from 0.6 to 0.8, while the wavefront sensing time for each image is reduced to just 30 ms. This approach provides a highly efficient and reliable solution for wavefront sensing of extended targets in dynamic environments.