6533b871fe1ef96bd12d25ec
RESEARCH PRODUCT
Optimization and sensitivity analysis of existing deep learning models for pavement surface monitoring using low-quality images
Ronald RobertsFabien MenantGaetano Di MinoVincent Baltazartsubject
Control and Systems EngineeringSettore ICAR/04 - Strade Ferrovie Ed AeroportiDeep learningMonitoring pavement surfacesBuilding and ConstructionCivil and Structural EngineeringTransfer learningPavement distressePavement management systemdescription
Automated pavement distress detection systems have become increasingly sought after by road agencies to in crease the efficiency of field surveys and reduce the likelihood of insufficient road condition data. However, many modern approaches are developed without practical testing using real-world scenarios. This paper ad dresses this by practically analyzing Deep Learning models to detect pavement distresses using French Secondary road surface images, given the issues of limited available road condition data in those networks. The study specifically explores several experimental and sensitivity-testing strategies using augmentation and hyper- parameter case studies to bolster practical model instrumentation and implementation. The tests achieve adequate distress detection performance and provide an understanding of how changing aspects of the workflow influence the actual engineering application, thus taking another step towards low-cost automation of aspects of the pavement management system
| year | journal | country | edition | language |
|---|---|---|---|---|
| 2022-08-01 |