Spatial relation extraction (SRE) aims to identify and classify spatial relations between entities in textual data. Previous methods often neglected the significance of element types, which is crucial for accurate spatial relation classification. To address this issue, we present a novel framework that integrates type correlation and structural constraints to enhance spatial relation extraction. To integrate element types into spatial relation extraction, we first construct virtual type words representing element types and then design an auxiliary task to predict element types. Additionally, we introduce structural loss during training to strengthen the connection between the auxiliary and main tasks while constraining the embedding structure of the virtual type words. Finally, we capture knowledge and structural information related to element types in relation classification using Type Correlation. Experimental results on the SpaceEval dataset demonstrate the superior performance of our model over state-of-the-art baselines, particularly in distinguishing complex relations and addressing challenges with null-role relations. These findings highlight the potential of leveraging element type correlations to advance spatial reasoning and natural language understanding.