The precision and real-time capability of monitoring the condition of electrical transformers are critical to ensuring the stable operation of power systems and enhancing the efficiency of power transmission. To improve both the accuracy and performance of electrical transformer condition monitoring, this paper proposes an enhanced DETR model, termed GLD-DETR, which incorporates the GOLD-YOLO method. The main contributions of our model are as follows: (1) Our model integrates the Re-param Block into an extended residual module within the backbone. This architecture enables the model to more accurately detect subtle changes in transformer characteristics across various power application scenarios, thereby enhancing the robustness of feature extraction. (2) We implement the Gather-and-Distribute (GD) mechanism, which allows the model to effectively utilize feature information derived from monitoring data. This mechanism facilitates the integration of state features from transformers at multiple scales, thereby preventing the loss of valuable information and ensuring comprehensive feature representation. (3) We introduce the novel application of the cascade attention mechanism, which diversifies attention by allocating different attention heads to various features. This enhancement improves the model’s ability to focus on the most critical state features, thereby enhancing the overall accuracy and reliability of transformer condition monitoring.