A Lightweight Traffic Sign Detection Algorithm based on Improved YOLOv7

In order to achieve a lightweight network for traffic sign detection, we made improvements to the YOLOv7 object detection network, which include the following: 1. Replacing the original backbone with a lightweight PP-LCNet; 2. Using the CARAFE upsampling operator instead of the conventional upsampling methods; 3. Incorporating the SimAM attention module into the network; 4. Decoupling the detection head of YOLOv7. The training and testing were conducted on the CCTSDB dataset. The experimental results demonstrate that the improved YOLOv7 in this study has fewer parameters and lower computational complexity while achieving higher accuracy on the CCTSDB dataset. These findings indicate that the proposed modifications effectively enhance the performance of YOLOv7 for traffic sign detection while maintaining a lightweight network architecture.