The Internet of Things (IoT) connects devices and systems but faces significant security challenges from increasing cyberattacks. Traditional intrusion detection systems rely on deep learning but suffer from limited training data availability. This research proposes an innovative system based on Deep Transfer Learning using a hybrid model combining Swin Transformers with LSTM networks. Swin Transformers excel at processing data hierarchically across multiple scales, while LSTM networks handle sequential dependencies in network traffic data. The proposed system was evaluated using benchmark datasets and achieved advanced results in detecting security threats in IoT environments, demonstrating the effectiveness of combining these techniques to enhance the security of intelligent systems.