This research paper proposes an adaptive traffic signal control method based on Proximal Policy Optimization (PPO) integrated with a Transformer architecture, utilizing exclusively RFID-based vehicle detection and type recognition (car, bus, ambulance) within the SUMO simulation environment. RFID readers function as vehicle detectors, generating a stream of aggregated counts in short time windows. This stream is sequentially modelled by the Transformer, enabling the PPO policy to capture inflow variability and determine phase maintenance or switching decisions while adhering to safety constraints. The reward function is designed to minimize global travel time and queue lengths. We experimentally compare the proposed method against a classic sequential algorithm and the adaptive Miller algorithm, using identical input data and phase constraints. Results indicate that the PPO-Transformer achieves a reduction in average delay by 28.6% and queue lengths by 36.0% compared to the fixed-time baseline. Furthermore, the model outperforms the adaptive Miller algorithm, reducing delays by 9.1% and the number of stops by 9.5%, while simultaneously increasing total intersection throughput by 12% (relative to the fixed-time baseline). Sensitivity analysis demonstrates the robustness of the PPO-Transformer to partial RFID read losses and bursty traffic inflows.