摘要
本文针对复杂路面灾害检测问题,提出了一种改进的YOLOv8n模型。通过引入FasterC2f模块替代原C2f模块减少参数量,通过优化损失函数降低低质量样本的不良梯度影响,并采用大型可分离核注意力机制增强小目标灾害特征表达能力。实验基于RDD2022数据集验证,改进算法F1值达0.827,较原YOLOv8n提升17.8%;m AP50/95指标提高至83.1,增长15.4%;模型参数量减少6%。所提算法在保持轻量化的同时显著提升了复杂路面多类型灾害的检测精度与效率。
This paper proposes an improved YOLOv8n model for complex road surface disaster detection.By introducing the FasterC2f module to replace the original C2f module,the parameter count is reduced.The loss function is optimized to mitigate adverse gradient effects from low-quality samples,and a large separable kernel attention mechanism is employed to enhance feature representation of small disaster targets.Experiments based on the RDD2022 dataset demonstrate that the improved algorithm achieves an F1 value of 0.827,a 17.8%increase compared to the original YOLOv8n;the mAP50/95 metric increases to 83.1,a growth of 15.4%;meanwhile,the model parameters are reduced by 6%.The proposed algorithm significantly improves the detection accuracy and efficiency of multiple types of disasters on complex road surfaces while maintaining a lightweight architecture.
作者
张维洋
ZHANG Weiyang(Xi'an Shiyou University,Xi'an Shaanxi 710065)
出处
《软件》
2025年第12期94-96,共3页
Software
关键词
YOLOV8
复杂路面
多类型灾害
YOLOv8
complex road surfaces
multiple types of disasters
