基于改进DeepCrack网络的山区地裂缝检测算法
Mountainous ground crack detection algorithm based on the improved DeepCrack network
黄海新 1郭鹏1
作者信息
- 1. 沈阳理工大学自动化与电气工程学院,辽宁沈阳 110159
- 折叠
摘要
针对山区地表裂缝检测存在背景复杂、裂缝和背景像素比例不均衡现象,以及传统裂缝检测算法的效果差,精度低、泛化性能不足等问题,提出采用DeepCrack网络作为基本构架,同时对基础网络进行改进,在网络中加入位置注意力机制,调整损失函数,并将优化器换为Adan优化器.为证明所提出算法的有效性和准确性,将DeepCrack数据集与人工标注的数据集结合,采用F-feature指标来评估检测性能.实验结果表明,改进算法在数据集相对原算法整体精度提升1.61%,因此改进算法提升了裂缝检测的准确性,具有较好的检测效果.
Abstract
Aiming at the problems of complex background,unbalanced ratio of cracks and background pixels in mountainous ar-eas,as well as the poor effect of traditional crack detection algorithms,low accuracy,and insufficient generalisation performance,the DeepCrack network is proposed as the basic architecture,while the basic network is improved by adding a Coordinate Attention mecha-nism,adjusting the loss function,and replacing the optimiser with the Adan optimiser.To demonstrate the effectiveness and accuracy of the proposed algorithm,the DeepCrack dataset is combined with the manually labelled dataset,and the F-feature metric is used to eval-uate the detection performance.The experimental results show that the overall accuracy of the improved algorithm in the dataset is im-proved by 1.61%relative to the original algorithm,so the improved algorithm enhances the accuracy of crack detection and has better detection results.
关键词
地表裂缝检测/DeepCrack网络/位置注意力机制/Adan优化器/F-feature指标Key words
Detection of surface cracks/DeepCrack network/Coordinate attention mechanism/Adan optimizer/F-feature引用本文复制引用
出版年
2024
国家科技期刊平台
NSTL
维普
万方数据