Direct current (DC) series arc fault detection is a critical challenge in electrical safety systems, which requires high accuracy, low latency, and power efficiency. Convolutional neural networks (CNNs) have shown promise in addressing this challenge, but their implementation on suitable embedded hardware platforms remains an active research area. This study presents an approach using CNNs accelerated on Field-Programmable Gate Arrays (FPGAs) for efficient arc fault detection. We leverage the STANN library for high-level synthesis (HLS) to overcome challenges associated with CNN implementations for FPGA acceleration. Our fine-tuned CNN model achieves high accuracy, which is maintained when implemented on the FPGA platform, while also delivering low latency. A comparative analysis with the Nvidia Jetson GPU platform demonstrates the performance advantages of our FPGA-based solution. This work contributes to the advancement of real-time, efficient arc fault detection systems, potentially enhancing the safety and reliability of DC electrical systems across various industries.