Infrared target detection is limited by strong background interference, weak texture information, and blurred structural features. Existing deep learning-based methods mostly rely on single-target features, neglecting inter-target correlation information. To address this issue, this paper proposes a graph convolutional network-based infrared target detection algorithm that integrates inter-class correlation and semantic similarity, with the core being the construction of a dynamically fused graph structure. Firstly, a static co-occurrence adjacency matrix is constructed using the co-occurrence probability of targets in the image to capture the contextual relationship between targets. Simultaneously, a semantic adjacency matrix is constructed based on pre-trained word vectors and updated during training to dynamically adapt to the semantic characteristics of target labels. Subsequently, the two adjacency matrices are fused, and the category word vectors are input into the graph convolutional network as node features, achieving high-level modeling of category relationship. Finally, the obtained relational features are fused with features extracted by the YOLO11 backbone network through an attention mechanism for subsequent classification and localization branches of the detection head. Experimental results on a self-made infrared aircraft dataset demonstrate that, compared with YOLO11, the proposed algorithm improves mAP50 and mAP50:95 by 1.08% and 0.86%, respectively, with comparable computational complexity and parameter count. Compared with other state-of-the-art algorithms, the proposed algorithm also achieves optimal detection accuracy. Theoretical analysis and experimental results demonstrate the effectiveness of the proposed algorithm in infrared target detection task under complex environments.