Association matching of geological objects with different sources and representation of their structures, attributes and semantic relationships by models is an important support for later tasks such as semantic query and clustering. In this paper, we propose a twin network geological survey spatial entities and text description information association matching model based on attention mechanism for the problems of semantic heterogeneity and expression differences between geological survey spatial entities and external text descriptions. First, the attribute information of geological survey spatial entities is converted into text paragraphs, and the text semantics of geological spatial entities is encoded with the basic granularity of sentence vectors; then the two types of text objects are mapped into a unified vector space and input to the twin network for feature learning, and finally the experimental evaluation of model performance is conducted on the constructed real dataset. The results demonstrate that the model can better represent the sentence semantic information of geological survey spatial entities, and its recognition F1 value is improved by 8.4 percentage points compared with the benchmark experiment, which is better than the selected comparison method.