TY  - EJOU
AU  - Lin, Feng-Cheng 
AU  - Lin, Jia-Yan 
AU  - Hung, Chun-Yu 
AU  - Wijaya,  

TI  - Hybrid Embedded System for Identifying Road Condition Defects Using Three Adapted Deep Learning Models
T2  - Computers, Materials \& Continua

PY  - 
VL  - 
IS  - 
SN  - 1546-2226

AB  - Road defects, such as cracks and potholes, pose significant challenges to urban infrastructure management and transportation safety. Existing automated detection methods often suffer from computational inefficiency, sensitivity to environmental conditions, and a lack of severity assessment capabilities. Addressing these limitations, this study presents an integrated deep learning framework optimized for embedded platforms, designed to provide comprehensive, real-time road defect detection, classification, and tracking. The proposed framework utilizes the proposed FAST-UNet for efficient segmentation, an enhanced YOLOv11_SDIDC for robust detection, MobileNetV3 with SimAM attention for multi-level severity classification, and ByteTrack for stable tracking, optimized with a hybrid loss function. Deployed on the NVIDIA Jetson AGX Orin platform using TensorRT acceleration, the system demonstrates competitive predictive performance and efficient inference on the RDD2022 dataset. Furthermore, integrated GPS functionality enables real-time geospatial mapping. This research contributes a scalable, efficient, and multi-functional solution for automated road defect management, enhancing intelligent transportation systems in smart cities.
KW  - Deep learning; road defect detection; MobileNetV3; YOLOv11_SDIDC; ByteTrack

DO  - 10.32604/cmc.2026.085098