With the development of modern multi-sensor tracking technology, trajectory fusion plays a crucial role in fields such as flight target tracking and air traffic control. However, traditional trajectory fusion methods often face challenges in accuracy, robustness, and real-time performance when dealing with complex and dynamic targets and multi-source information. This paper proposes a trajectory fusion method based on fully convolutional dynamic difference for multi radar collaborative systems. The method combines fully convolutional network (FCN) and dynamic difference mechanism. The fully convolutional network can effectively handle large-scale input data, while the dynamic difference mechanism dynamically adjusts the fusion strategy by calculating the differences in sensor data in real time, overcoming the shortcomings caused by static models in traditional methods. Experiments have shown that in a multi radar system environment, the accuracy of trajectory fusion using this method is significantly better than traditional fusion methods, and the system can maintain high real-time and processing capabilities in complex noise environments.