<p>Deep learning algorithms have been successfully employed in the recognition of geochemical anomalies associated with mineralization. These deep learning-based geochemical anomaly recognition methods can be categorized into supervised and unsupervised learning approaches. Supervised learning methods typically require a substantial amount of labeled data and often exhibit limited generalization capabilities; therefore, they tend to perform poorly in untrained study areas. By contrast, unsupervised learning methods do not require labeled data for training and exhibit strong generalization capabilities, but some of the identified geochemical anomalies are not linked to mineralization. To address these issues, this study proposes a transfer-learning framework for geochemical anomaly recognition. This framework integrates the advantages of both unsupervised and supervised learning methods, including the generalization capabilities of the unsupervised methods and precise anomaly recognition abilities of the supervised methods. To further enhance the performance of this framework, a geological embedding layer was designed to facilitate the transfer of geological knowledge acquired from the source area to the target area, thereby improving geological knowledge transfer. A case study of a geochemical survey dataset from Hubei Province, China, was conducted to evaluate the effectiveness of the proposed framework. The western and southeastern regions of Hubei were selected as the source and target areas, respectively. The results demonstrated that the proposed transfer-learning framework could effectively delineate geochemically anomalous areas related to gold polymetallic mineralization in the target area, even if it had not been trained in this area. Furthermore, the incorporation of geological embedding layers for geological knowledge transfer notably enhanced model performance.</p>