High-resolution processing of seismic signals is crucial for enhancing the ability to characterize underground geological structures in detail and improving the accuracy of thin-layer reservoir identification. Traditional seismic high-resolution algorithms suffer from issues such as poor robustness, low computational efficiency, and neglect of inter-channel structural relationships. Meanwhile, most mainstream deep learning-based high-resolution methods rely on end-to-end networks, lacking guidance from prior information and ignoring differences between data domains, resulting in insufficient generalization capabilities. Therefore, this paper proposes a domain-adaptive knowledge distillation deep learning network for seismic data, DAKD-Net (Domain-Adaptive Knowledge Distillation Network). This method establishes a teacher-student network model using forward simulation datasets as training samples. The teacher network module establishes physical constraints between low- and high-resolution data, extracting high-frequency prior information during the guidance phase to direct the student network module in restoring details without prior conditions. Domain adaptation then generalizes the model to real seismic data, enhancing its generalization capability and structural representation accuracy in actual work area data. Structurally, DAKD-Net adopts a U-net backbone to fully extract spatial structural information across multi-track seismic profiles. Its training mechanism enables prior knowledge transfer through the teacher-student network, allowing high-resolution data recovery without prior information. In application, domain adaptation fine-tuning strategies enhance the network's generalization capability and structural representation effectiveness in real-world work areas. Experimental results demonstrate that the proposed method outperforms traditional approaches and classical deep networks in both vertical resolution and complex structural detail recovery, exhibiting robust performance and practical applicability.

Research on reservoir-unit division in fault-controlled oil and gas reservoirs is essential for analyzing reservoir hydrocarbon migration and accumulation. Currently, most research on reservoir-unit division has focused solely on the identification of faults and caves, employing three-dimensional spatial visualization or other methods for a simple analysis of their links. However, these approaches often lack a reasoning process that exploits the links between faults and caves for deeper insights. For such complex oil and gas reservoirs, a systematic analysis based on the interrelations between multiple geological factors is needed. Therefore, this paper proposes a graph-based method for reservoir-unit division in fault-controlled oil and gas reservoirs, enabling the representation of links between faults and caves, and it presents further systematic analysis to derive the reservoir-unit division results. A multi-attribute graph-clustering-based fault-extraction method is utilized to achieve comprehensive fault representations as fault entities. More reliable cave-instance segmentation results are obtained through attribute fusion, representing cavity entities. A graph incorporating fault and cave entities is then created. Fault entities are classified into several levels according to their spatial scale, and directed edges are utilized to represent connectivity links between faults and caves. Moreover, a connectivity analysis centered on caves was conducted using the created graph. Based on existing reservoir-unit knowledge and the cave-connectivity analysis results, reservoir-unit division was achieved. The proposed method provided reservoir-unit division results highly consistent with the information contained in seismic data, offering a new perspective for multielement integrated analysis in geophysical exploration.