Multimodal data fusion is a crucial method for power transmission line state perception and judgment. However, the influence of environmental factors on monitoring data quality and inter-data relationships can lead to deviations in the fusion process and judgment results. To address this question, we propose an enhanced state perception method for power transmission lines based on a multimodal data semantic topology graph. This method incorporates environmental semantic information during the multimodal data fusion process, thereby improving the accuracy of state decision-making. First, a scene segmentation algorithm for the transmission line corridor divides the monitoring range into regions with distinct features. Next, data sources are filtered, and a semantic topology graph of the data is constructed based on a spatial proximity knowledge base. Subsequently, targeted neural networks are employed for feature extraction from the multimodal data. Finally, feature fusion is performed using a graph attention network and conditional random fields, and Softmax is utilized for state perception decision-making. Simulation experiments demonstrate the efficacy of the proposed method in the state perception and decision-making of transmission lines, achieving a state detection accuracy of 92% and thereby providing a robust assurance for comprehensive transmission line state monitoring.