中国计量大学学报2023，Vol.34Issue(4)：511-518,8.DOI:10.3969/j.issn.2096-2835.2023.04.003
基于MobileNeXt与多尺度区域分割的激光对位检测
Laser alignment detection based on MobileNeXt and multi-scale region segmentation
摘要
Abstract
Aims:This paper aims to improve the assembly process accuracy in the manufacturing process of laser ranging equipment.Methods:A laser alignment detection algorithm was proposed based on MobileNeXt and multi-scale region segmentation to optimize the production process of laser ranging equipment.In the alignment detection stage,MobileNeXt was used as the backbone network to compute image features at low cost.And depth separable coordinate attention was introduced into the backbone network to enhance the model's ability to capture effective features.The network was trained using a knowledge evolution training scheme to obtain a lightweight and high-precision classification model.In the validation stage,a laser center alignment verification algorithm with multi-scale region segmentation was designed.The algorithm first processed the original image through linear transformation and bilateral filtering.It then segmented the image into multiple scales and extracted features in different regions.Finally,a threshold was designed based on the laser alignment feature to verify the correctness of the alignment result.Results:The algorithm achieved a center alignment detection accuracy of 99.97％and an alignment error of 37.96 μm.Conclusions:The proposed algorithm meets the requirements of practical laser alignment detection tasks.关键词
对位检测/轻量化网络/灰度分割/注意力机制Key words
alignment detection/lightweight neural network/gray segmentation/attention mechanism分类
信息技术与安全科学引用本文复制引用
丁若男,倪军,陈家伟,樊其明,龚华平..基于MobileNeXt与多尺度区域分割的激光对位检测[J].中国计量大学学报,2023,34(4):511-518,8.基金项目
国家重点研发计划课题(No.2022YFB3903204) （No.2022YFB3903204）
