基于多层级特征融合的公路病害检测模型研究  

Research on Highway Disease Detection Model Utilizing Multi-level Feature Fusion Techniques

作　　者：贾睿 孟祥诚 徐春雨[1] JIA Rui;MENG Xiangcheng;XU Chunyu

机构地区：[1]辽宁省交通高等专科学校,辽宁沈阳110122

出　　处：《辽宁省交通高等专科学校学报》2025年第1期1-4,共4页Journal of Liaoning Provincial College of Communications

基　　金：辽宁省交通高等专科学校技术应用型科研项目(xqnlt202302);辽宁省教育科学规划课题(JG22EB252)。

摘　　要：为实现智能准确地检测公路病害,提升公路智能养护水平,本文提出一种基于多层级特征融合的公路病害检测模型HD-Net(Highway Defects Network)。此模型设计多层级特征融合模块,融合浅层细节特征和高层抽象语义特征,提高全局的特征表达能力,实现对多种复杂环境的公路病害的准确识别。实验结果表明,HD-Net检测模型在不同场景中取得优秀结果,表明多层级特征的有效融合能够提升模型在复杂环境中的公路病害检测能力。Aiming to achieve intelligent and precise detection of highway defects and elevate the benchmark for intelligent highway maintenance,this paper introduces the HD-Net(Highway Defects Network).HD-Net incorporates multi-level feature fusion module which integrates low-level detailed features with high-level abstract semantic features.This integration amplifies the overall feature expression capability,allowing for accurate recognition of highway diseases in a multitude of complex environments.Experimental results validate that the HD-Net model exhibits exceptional performance across diverse scenarios,confirming that effective multi-level feature fusion substantially enhances the model's capacity to detect road diseases in challenging settings.

关 键 词：公路病害检测 多层级特征融合 注意力机制 

分 类 号：U418.6[交通运输工程—道路与铁道工程]