Research on highway road condition intelligent assessment and optimization system based on deep learning and internet of things
With the acceleration of urbanization and the continuous increase in car ownership, the real-time evaluation and optimization of highway road conditions, as an essential component of modern transportation systems, has become an urgent issue in traffic management. Experimental verification shows that...
Saved in:
| Main Authors: | , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Elsevier
2025-12-01
|
| Series: | Systems and Soft Computing |
| Subjects: | |
| Online Access: | http://www.sciencedirect.com/science/article/pii/S2772941925001796 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | With the acceleration of urbanization and the continuous increase in car ownership, the real-time evaluation and optimization of highway road conditions, as an essential component of modern transportation systems, has become an urgent issue in traffic management. Experimental verification shows that the model significantly improves prediction performance. One of the core components of an intelligent highway road condition assessment system is the ability to accurately detect and classify objects in real-time, such as potholes, cracks, debris, and traffic signs. Two prominent object detection frameworks, RetinaNet and YOLO, have been widely adopted in computer vision tasks due to their complementary strengths. RetinaNet, a state-of-the-art two-stage detection algorithm, leverages Focal Loss to address the challenge of class imbalance, making it particularly effective in detecting rare and hard-to-classify objects. By optimizing the ResNet-50 down sampling module, introducing the channel attention mechanism, and improving the NMS strategy, the detection accuracy is significantly improved to 89.63 % mAP while maintaining efficient processing speed. The enhanced optimized vehicle detector introduced in this article has high real-time performance. Customizing prior information, enhancing Mixup data, improving LabelSmoothing to enhance generalization ability, and optimizing GIoU position loss and FocalLoss confidence loss are combined with the CBAM(Cost Benefit Analysis Method) module to improve network structure. In the end, mAP significantly increased to 88.35 %, and the detection speed remained at 31.14FPS. |
|---|---|
| ISSN: | 2772-9419 |