To address the challenges of strong noise interference, pronounced non-stationary characteristics, and difficulties in feature representation of arc current signals in high-speed railway pantograph-catenary systems, this article proposes a dual-branch pantograph-catenary arc detection method that integrates a structure-aware generative adversarial network (SA-GAN), a receptive field attention residual network (RFA-ResNet50), and a Transformer. First, an improved cuckoo catfish optimization (ICCO) algorithm is usd to adaptively optimize the parameters of the Kalman filter, thereby denoising arc signals while suppressing background noise and preserving abrupt arc features. Subsequently, one-dimensional time-series signals are transformed into two-dimensional images using the Markov transition field (MTF), and a SA-GAN is introduced to enhance structural information, thereby improving the discriminability and representational capability of critical features. During feature extraction, a spatial-temporal dual-branch network is constructed. The SA-GAN-enhanced RFA-ResNet50 branch extracts local spatial features through multi-scale convolutions and a receptive field attention mechanism, while the Transformer branch captures long-range temporal dependencies using self-attention to achieve global modeling of dynamic arc evolution. A feature fusion strategy is then employed to facilitate the collaborative representation of spatial and temporal information. In addition, transfer learning is incorporated to mitigate the issue of insufficient training under limited-sample conditions and improve model generalization. Finally, comprehensive performance evaluations are conducted on experimental datasets collected under four operating conditions and compared with several representative baseline models. The experimental results show that the proposed method achieves superior performance in signal-to-noise ratio enhancement, feature representation capability, and detection accuracy, enabling high-precision identification of pantograph-catenary arcs under complex operating conditions.