基于结构建模与查询机制的道路病害识别与定位方法OA
Road disease identification and localization method based on structural modeling and query mechanism
针对道路病害图像分类与检测任务中局部关系建模不足、背景干扰强及多尺度适应性差等问题,提出一种基于结构建模与查询机制的识别定位方法.分类中结合区域划分、GCN与BiLSTM增强局部结构表达;检测模块中引入多尺度查询预测分支提升复杂场景定位能力,并采用Focal Loss与温度缩放优化类别不平衡及置信度校准.在RDD数据集上,该方法取得了较优的分类与检测性能.
To address the problems of insufficient local relationship modeling,strong background interfer-ence,and poor multi-scale adaptability in road distress image classification and detection tasks,this paper proposed a recognition and localization method based on structural modeling and a query mechanism.For classification,region partitioning,a graph convolutional network(GCN),and a bidirectional long short-term memory(BiLSTM)network are integrated to enhance the representation of local structural patterns.For the detection module,a multi-scale query prediction branch is introduced to improve localization capability in complex scenes,while Focal Loss and temperature scaling are employed to alleviate class imbalance and optimize confidence calibration.Extensive experiments on the RDD data set demonstrate that the proposed method achieves favorable classification and detection performance.
钟昆志;王晓筱;许兵;项博涛;肖倩
江西省交通投资集团有限责任公司,江西南昌 330108南通大学交通与土木工程学院,江苏南通 226019江西省交通投资集团有限责任公司,江西南昌 330108交通运输部公路科学研究院,北京 100088交通运输部公路科学研究院,北京 100088
交通工程
道路病害图像分类目标检测图卷积网络查询式检测置信度校准
road distressimage classificationobject detectiongraph convolutional network(GCN)query-based detectionconfidence calibration
《长春工程学院学报(自然科学版)》 2026 (3)
39-48,10
江西省交通运输厅科技项目(2025ZG004)
评论