This paper proposes an intelligent traffic signal control system that combines multimodal data fusion and deep reinforcement learning to optimize traffic flow through automated signal timing adjustments. The system integrates multiple data sources, including camera video, radar data, GPS tracking and other data sources, to create a comprehensive traffic state representation. Specifically, we design a multimodal fusion network where Convolutional Neural networks process image data, while Long Short-term Memory networks handle time series data. The image and time series features extracted by their respective network are then weighted and fused using an attention mechanism. Subsequently, the fused traffic state representation is fed into a Deep Q-Network, aiming to learn the optimal signal control policies through reinforcement learning. These developments will play a key role in advancing intelligent traffic control systems, ensuring smoother traffic flow, reducing delays, and enhancing both the safety and efficiency of urban transportation networks.