(1) Background: Geological surveying is undergoing a digital transformation process towards the adoption of intelligent methods in China. Cognitive intelligence methods, such as those based on knowledge graphs and machine reading, have made progress in many domains and also provide a technical basis for quality detection in unstructured lithographic description texts. (2) Methods: First, the named entities and the relations of the domain-specific knowledge graph of petrography were defined based on the petrographic theory. Second, research was carried out based on a manually annotated corpus of petrographic description. The extraction of N-ary and single-entity overlapping relations and the separation of complex entities are key steps in this process. Third, a petrographic knowledge graph was formulated based on prior knowledge. Finally, the consistency between knowledge triples extracted from the corpus and the petrographic knowledge graph was calculated. The 1:50,000 sheet of Fengxiangyi located in the Dabie orogenic belt was selected for the empirical research. (3) Results: Using machine reading and the knowledge graph, petrographic knowledge can be extracted and the knowledge consistency calculation can quickly detect description errors about textures, structures and mineral components in petrographic description. (4) Conclusions: The proposed framework can be used to realise the intelligent inspection of petrographic knowledge with complex entities and relations and to improve the quality of petrographic description texts effectively.

Lithological recognition is a fundamental task and represents a foundational application domain for artificial intelligence (AI) in geosciences. Although contemporary data-driven methods excel at extracting features from rock images, they often exhibit limited generalisability, and their predictive confidence rankings frequently misalign with established geological priors. Inspired by the integration of governing physical equations into the loss function and the associated optimisation framework in physics-informed neural networks (PINNs), this study proposes the PetroSpear learning framework. This novel approach embeds semantic similarity priors of rock classes directly into the image classification pipeline, thereby realising a data- and knowledge-driven model that enhances both predictive accuracy and generalisability. Specifically, we first derive pairwise similarities between rock classes using geological prior knowledge to construct a similarity matrix. Subsequently, we formulate a semantic prior embedding and ranking (Spear) loss function that incorporates this matrix to constrain the semantic ordering. To maintain differentiability, the loss computation transitions from standard probability distributions to the optimisation of a differentiable rank correlation metric between the predicted and ground-truth class rankings. Furthermore, PetroSpear employs a two-stage training strategy: the initial stage leverages the rapid convergence of the standard cross-entropy (CE) loss to establish foundational decision boundaries, while the subsequent stage jointly optimises both the CE and Spear losses to mitigate the poor generalisation and prior misalignment characteristic of purely data-driven models. Experimental results demonstrate that the proposed approach significantly outperforms baseline models trained exclusively with the CE loss. Notably, this study reveals that the Spear loss weight does not exhibit a monotonic positive correlation with model performance, underscoring the necessity of balancing empirical data fitting with integration to geological prior knowledge. Models leveraging domain-specific word embedding and domain knowledge graph embedding consistently outperform general-purpose word embedding models. These findings further demonstrate that, despite the rapid advancements of artificial general intelligence (AGI), research on geoscience knowledge representation retains irreplaceable value.