| 引用本文: | 朱成杰,蔡子正,朱洪波.基于改进 YOLOv8 的路面坑洼检测(J/M/D/N,J:杂志，M：书，D：论文，N：报纸).期刊名称,2026，43（2）：93-100 |
| CHEN X. Adap tive slidingmode contr ol for discrete2ti me multi2inputmulti2 out put systems[ J ]. Aut omatica, 2006, 42(6): 4272-435 |
|
| 摘要: |
| :目的 针对现有路面坑洼检测模型参数量大,检测实时性较差,路面环境复杂而导致检测准确度不高的问
题,提出一种基于改进 YOLOv8 的路面坑洼检测算法。 方法 首先,将颈部网络替换为双向金字塔网络 Bifpn(Bidirectional Feature Pyramid Network),进行不同尺度的特征融合;其次,为提高模型推理速度,使用多分支结构 DBB
(Diverse Branch Block)替换 C2f 模块中的卷积层,提出了重参数化模块 C2f-DBB 并替换部分 C2f 模块;然后,在骨
干网络末端引入通道先验注意力机制 CPCA(Channel Prior Convolutional Attention),在有效提取空间关系的同时保
留通道先验,强化特征提取能力;最后,使用边界框回归损失函数 MPDIoU(MPD Intersection over Union)代替原模
型的 CIoU(Complete Intersection over Union),进一步提高算法精度。 结果 实验结果表明:改进后的算法在平均精度
上较原始网络提高了 3. 1%,而模型参数量仅为 5. 9 M,计算量为 7. 3 G,分别下降了 6. 3%和 9. 8%,每秒帧数可达
53. 4。 结论 通过实验结果可以看出:改进后的模型与当前主流模型相比在精度上提升显著,且满足实时性要求,对
于实际应用具备一定的参考价值。 |
| 关键词: 坑洼检测 YOLOv8 重参数化 注意力机制 |
| DOI： |
| 分类号: |
| 基金项目: |
|
| Road Surface Pothole Detection Based on Improved YOLOv8 |
|
ZHU Chengjie CAI Zizheng ZHU Hongbo
|
|
School of Electrical and Information Engineering Anhui University of Science and Technology Huainan 232001 Anhui
China
|
| Abstract: |
| Objective To address the issues of large model parameters poor real-time detection performance and low
detection accuracy due to complex road environments in existing road surface pothole detection models an improved
YOLOv8-based road surface pothole detection algorithm is proposed. Methods First the neck network was replaced with
a bi-directional feature pyramid network BiFPN to achieve multi-scale feature fusion. Second to enhance the inference
speed of the model a diverse branch block DBB was used to replace the convolutional layers in the C2f module and a
reparameterization module C2f-DBB was introduced to substitute part of the C2f modules. Then a channel prior
convolutional attention CPCA mechanism was introduced at the end of the backbone network to retain channel priors
while effectively extracting spatial relationships thereby strengthening feature extraction capabilities. Finally the
boundary box regression loss function MPD intersection over union MPDIoU was used instead of the original model?? s
complete intersection over union CIoU further improving the accuracy of the algorithm. Results Experimental results
showed that the improved algorithm increased the average precision by 3. 1% compared to the original network with the
model parameters being only 5. 9 M and the computation being 7. 3 G representing reductions of 6. 3% and 9. 8%
respectively. The frame rate reached 53. 4 frames per second. Conclusion The experimental results indicate that the improved model significantly enhances accuracy compared to current mainstream models while meeting real-time
requirements providing valuable reference for practical applications. |
| Key words: pothole detection YOLOv8 reparameterization attention mechanism |
