Research on Road Defect Detection Method Based on Adaptive Perception Module Improving YOLOv8

DOI: https://doi.org/10.62517/jbdc.202601309

Author(s)

Jianze Liu

Affiliation(s)

School of Mathematics and Statistics, Beihua University, Jilin, Jilin, China *Corresponding Author.

Abstract

Road defect detection is an important part of road maintenance management. In actual road images, defect targets such as cracks and potholes often have problems such as weak edges, unclear textures, and irregular shapes. At the same time, they are easily interfered with by background factors such as shadows, markings, stains, and repair traces, which makes the detection model prone to missing detections and false detections. To address these issues, this paper proposes a YOLOv8 road defect detection method that integrates an adaptive perception module. This method is based on the YOLOv8x detection framework and introduces an Adaptive Perception Module (APM) into the network. Through convolutional feature enhancement, residual connections, and attention weighting mechanisms, it improves the model's perception ability of the edge, texture, and local structure information of the diseased area. To analyse the impact of the embedding position of APM on the detection performance, this paper embeds APM at the Backbone P4, Backbone P3, Neck P4, and Neck P3 positions respectively, and conducts comparative experiments with the baseline model that does not incorporate APM. The experimental results show that the detection gain of the APM module exhibits significant location dependence. Among them, the APM-Neck-P4 model achieves the best overall performance, with precision, recall, F1, mAP@0.5, and mAP@0.5:0.95 being 0.5760, 0.4658, 0.5151, 0.4836, and 0.2416, respectively. Compared with the baseline, the mAP@0.5 of APM-Neck-P4 increases by 0.0316, and the mAP@0.5:0.95 increases by 0.0271, indicating that introducing APM at the Neck P4 position can effectively enhance the multi-scale fusion feature representation and improve the accuracy of road defect detection. The research results can provide a reference for the structural design and embedding position selection of the feature enhancement module in road defect detection.

Keywords

Road defect Detection; YOLOv8; Adaptive Perception Module; Feature Enhancement; Object Detection

References

[1] Yi, M. (2025). Research on detection and maintenance methods for major highway diseases. MS, 15(01), 115–123. https://doi.org/10.12677/ms.2025.151014 [2] Sun, L., Liu, Z., & Xie, Z. (2024). Road crack detection in orthophotos based on EC-YOLOX algorithm. Preprints. https://doi.org/10.20944/preprints202403.0680.v1 [3] Jiang, Y. (2024). Road damage detection and classification using deep neural networks. Discover Applied Sciences, 6(8), 421. https://doi.org/10.1007/s42452-024-06129-0 [4] Maeda, H., Kashiyama, T., Sekimoto, Y., Seto, T., & Omata, H. (2021). Generative adversarial networks for road damage detection. Computer-Aided Civil and Infrastructure Engineering, 36(1), 47–60. https://doi.org/10.1111/mice.12561 [5] Feng, X. (2021). A survey on the development of image data augmentation. Computer Science and Application, 11(02), 370–382. https://doi.org/10.12677/CSA.2021.112037 [6] Li, A., Ge, Y., Liu, H., Yang, C., & Zhou, X. (2024). Content-aware interpretable pavement damage detection model. Journal of Computer Research and Development, 61(3), 701–715.