摘要
针对深度学习算法检测钢材表面缺陷时,结构信息减少导致检测精度低的问题,提出一种特征融合和级联检测网络的Faster R-CNN钢材表面缺陷检测算法。首先利用主干网络提取特征图,通过融合特征图的方式,达到减少结构信息丢失的目的;进一步将生成的特征图输入RPN网络生成区域建议框;最后利用检测网络对区域建议框进行分类与回归,通过级联2个检测网络,实现精确检测结果的目标。对模型进行对比性实验分析,找出检测精度最优的算法模型。在NEU-DET数据集上对提出的算法进行了检验,主干网络采用VGG-16比采用Resnet-50的检测精度提高了2.40%;通过融合特征,检测精度提高了11.86%;通过检测网络的级联,检测精度提高了2.37%.通过对算法模型的不断改进和优化,检测精度达到了98.29%.与传统的钢材表面检测方法相比,改进算法能够更准确地检测出钢材表面缺陷的种类和位置,提升对钢材表面缺陷的检测精度。
A Faster R-CNN steel surface defect detection algorithm based on feature fusion and cascade detection network was proposed to solve the problem of low detection accuracy caused by reduced structure information when Deep Learning algorithm was used to detect steel surface defects.The improved Faster R-CNN algorithm is used to detect surface defects of steel.First,the feature map is extracted from the main network and fused to reduce the loss of structural information;Then the resulting feature map is further input into the RPN network generation area recommendation box;Finally,the detection network is used to classify and regress the regional recommendation box,and two detection networks are cascaded to achieve the target of accurate detection results.The model was analyzed by comparative experiments to find the algorithm model with the best detection accuracy.The proposed algorithm was tested on the NEU-DET dataset.The detection mean average precision of the backbone network using VGG-16 is 2.40%higher than that using Resnet-50.By fusing the features,the detection mean average precisionis improved by 11.86%.By detecting the cascade of the network,the detection mean average precision is improved by 2.37%.By continuously improving and optimizing the algorithm model,the detection mean average precision reaches 98.29%.Compared with traditional steel surface detection methods,this algorithm can detect the types and locations of steel surface defects more accurately,and improve the detection accuracy of steel surface defects.
作者
韩强
张喆
续欣莹
谢新林
HAN Qiang;ZHANG Zhe;XU Xinying;XIE Xinlin(College of Electrical and Power Engineering, Taiyuan University of Technology, Taiyuan 030024, China;Key Lab of Advanced Control and Intelligent Equipment of Shanxi, Taiyuan University of Science and Technology, Taiyuan 030024, China)
出处
《太原理工大学学报》
CAS
北大核心
2021年第5期754-763,共10页
Journal of Taiyuan University of Technology
基金
山西省自然科学基金资助项目(201801D121144,201901D211079)
先进控制与装备智能化山西省重点实验室开放课题基金资助(ACEI202101)。