<p>Seismic exploration is currently the most mature approach for investigating subsurface structures, yet the random noise greatly restricts its imaging accuracy. Previous methods face significant challenges: traditional computational methods are often computationally complex and their effectiveness is hard to guarantee; deep learning methods rely heavily on datasets, and the complexity of network training makes them difficult to apply in practical field scenarios. In this paper, we propose an unsupervised adaptive convolutional filtering (ACF) method based on a lightweight deep learning method. It is a lightweight adaptive denoising model with only 2,464 learnable parameters, representing a substantial reduction compared with mainstream deep learning networks. And ACF operates in a dataset-free manner, optimizing its parameters relying purely on internal data priors rather than external training data. We propose two types of priors: the local prior and the global variance prior for unsupervised learning, and put forward low-scale learning to further enhance its performance in noise processing. We validated our method on both 2D and 3D synthetic and field data, and the results demonstrate that ACF offers a favorable balance between noise attenuation and signal preservation, providing a competitive alternative to conventional and deep learning-based methods, especially in complex field scenarios.</p>