Ziyi Yang

To enable the deployment of road distress detection models on resource-constrained embedded platforms, this paper presents a compression case study based on the LRDD-YOLOv8n detector designed for real-time 1080p video input. Layer-adaptive magnitude-based pruning (LAMP) was integrated with channel-wise knowledge distillation. First, LAMP performs structured pruning adaptive global sparsity allocation to reduce parameters and FLOPs. Then, a larger teacher model (LRDD-YOLOv8s) with high structural similarity guides the pruned student to recover feature representations. Compared to the baseline LRDD-YOLOv8n (64.4% mAP@0.5, 2.02 M parameters, 5.9G FLOPs, and 55.5 ms GPU inference time on Jetson TX2), our compressed model under a 1/1.4 target compression ratio achieves a mAP@0.5 of 65.1% (an slight accuracy increment of 0.7%), while reducing parameters by 36.1% (to 1.29 M) and FLOPs by 30.5% (to 4.1 G). Deployed on the BOXER-8120AI edge platform (Jetson TX2), the optimized model achieves an average inference time of 48.3 ms per frame (a 13.0% latency reduction compared to the baseline model). In addition, a 20 FPS detection rate was sustained under the 30 FPS maximum hardware acquisition limit of the industrial camera stream. Kinematic and geometric analysis validates that this processing rate utilizes 66.7% of all physically available frames and establishes a 95.4% consecutive frame-to-frame spatial overlap at typical inspection vehicle speeds (40–60 km/h).

Automated pavement distress detection (PDD) is critical for the structural health monitoring (SHM) of transportation infrastructure, yet existing methods struggle with real-time multi-target detection under resource constraints. In this paper, YOLOv8-PDD was constructed based on YOLOv8 by introducing the large separable kernel attention (LSKA) mechanism module into the Spatial Pyramid Pooling—Fast (SPPF) module, replacing Complete-IoU (CIoU) loss with Distance-IoU (DIOU) loss as the loss function, and adopting Soft-Non-Maximum Suppression (NMS) to replace the original NMS algorithm. The proposed YOLOv8-PDD achieved 78.3% mean average precision with intersection over union above 0.5 (mAP@0.5 +8.1%) with a minimal complexity increase of +0.2 GFLOPs compared to the baseline YOLOv8n model. While incurring a negligible increase in latency (+0.09 ms), YOLOv8-PDD significantly outperforms YOLOv8n in detection accuracy (mAP@0.5 +8.1%), offering a superior accuracy–efficiency trade-off for real-time applications. YOLOv8-PDD performed well in detecting all categories, with AP values above 75% except for transverse crack and strip patch. Significant improvements in pothole detection AP@0.5 (+22.1%) and strip patch detection AP@0.5 (+17.7%) indicate superior small target and complex background adaptability. Our model achieved a detection efficiency of 68 frames per second (FPS) on consumer-grade CPUs (OpenVINO-optimized), outperforming 10 models (e.g., YOLOv5n and RTDETR-l) in accuracy–speed balance.

High-precision road distress detection (RDD) systems often face challenges in balancing accuracy, efficiency, and generalisation while managing high operational costs. To address these issues, this study proposed a lightweight RDD you only look once-based model (LRDD-YOLOv8) optimised for practical deployment. Key innovations include the C2f-faster module integrated into the neck network, the LS-detect head utilizing group normalisation and shared convolution to minimise computational overhead, and the Wise-minimum points distance intersection over union (MPDIoU) loss combines Wise-IoU and MPDIoU accelerating convergence. LRDD-YOLOv8 reduced parameters by 32.9% (from 3.01 M to 2.02 M) and floating operations by 28.4% (from 8.1 G to 5.8 G) compared to YOLOv8n, while improving mAP@0.5 by 1.3% (64.4% vs. 63.1%) on street view datasets deployed on the SenseTime BOXER-8120AI edge computing platform, LRDD-YOLOv8 achieved a 34.2% faster inference speed (55.5 ms vs. 84.3 ms per image) with maintained accuracy, demonstrating suitability for real-time applications. The model exhibited robust generalisation in complex backgrounds and effectively suppressed duplicate detections. These advancements position LRDD-YOLOv8 as a cost-effective solution for resource-constrained road maintenance tasks. Future efforts will focus on expanding distress categories and improving illumination-invariant feature learning to enhance adaptability.

Established unmanned aerial vehicle (UAV) highway distress detection (HDD) faces the dual challenges of accuracy and efficiency, this paper conducted a comparative study on the application of the YOLO (You Only Look Once) series of algorithms in UAV-based HDD to provide a reference for the selection of models. YOLOv5-l and v9-c achieved the highest detection accuracy, with YOLOv5-l performing well in mean and classification detection precision and recall, while YOLOv9-c showed poor performance in these aspects. In terms of detection efficiency, YOLOv10-n, v7-t, and v11-n achieved the highest levels, while YOLOv5-n, v8-n, and v10-n had the smallest model sizes. Notably, YOLOv11-n was the best-performing model in terms of combined detection efficiency, model size, and computational complexity, making it a promising candidate for embedded real-time HDD. YOLOv5-s and v11-s were found to balance detection accuracy and model lightweightness, although their efficiency was only average. When comparing t/n and l/c versions, the changes in the backbone network of YOLOv9 had the greatest impact on detection accuracy, followed by the network depth_multiple and width_multiple of YOLOv5. The relative compression degrees of YOLOv5-n and YOLOv8-n were the highest, and v9-t achieved the greatest efficiency improvement in UAV HDD, followed by YOLOv10-n and v11-n.