Abstract: Seismic impedance inversion is one of the essential technologies for reservoir prediction. In recent years, artificial intelligence has demonstrated promising application potential in the field of seismic impedance inversion. However, the lack of large-scale and high-quality labels constrains the accuracy and applicability of seismic impedance inversion methods in the framework of supervised learning, particularly when characterizing complex reservoir with features of rapid lithological changes, strong heterogeneity, and thin layers. In this paper, we proposed a seismic impedance inversion method based on hybrid attention mechanisms with convolutional block attention module(CBAM). The method constructs an intelligent seismic impedance inversion model in a self-supervised learning framework. The inversion model can be trained by the training data without labels, which can overcome the limitations imposed by insufficient labeled training data in seismic inversion applications. CBAM can effectively enhance the inversion model’ ability to extract critical information, improve the accuracy and reliability of seismic inversion, and reduce the dependence on the size of the training datasets. Application on the forward modeling data and field data demonstrate that the proposed method can achieve high-accuracy seismic impedance inversion and reservoir prediction for complex reservoirs.