M3E-Yolo: A New Lightweight Network for Traffic Sign Recognition
Traffic sign recognition is committed to ensuring the safety of automatic driving. Inspired by YOLOv5, this paper proposes a new model to solve the problem of poor balance between the accuracy and efficiency of existing algorithms in traffic sign recognition. Firstly, the lightweight network MobileNetV3 is introduced for feature extraction to reduce the number of parameters. Secondly, attention mechanism module is introduced to enhance channel features, which makes up for the reduced accuracy caused by the simplified model. Experiments show that the mAP value trained by our model on the Chinese traffic sign dataset reaches 93.6%, which is similar to the level of YOLOv5, and the number of parameters is less than a quarter of YOLOv5.
PDF
Datasets
| Task | Dataset | Model | Metric Name | Metric Value | Global Rank | Benchmark |
|---|---|---|---|---|---|---|
| Traffic Sign Detection | CCTSDB2021 | M3E-Yolo | mAP@0.5 | 93.4 | # 1 |
