基于ResNext-Vit的手绘工程图样的检测方法  

Detection Method of Hand Drawn Engineering Drawings Based on ResNext-Vit

作　　者：刘宪邦 栾天昕 林浩伟 詹瑞典 LIU Xianbang;LUAN Tianxin;LIN Haowei;ZHAN Ruidian(School of Advanced Manufacturing,Guangdong University of Technology,Guangzhou 510006,China)

机构地区：[1]广东工业大学先进制造学院,广东广州510006

出　　处：《机械工程与自动化》2025年第2期71-72,共2页Mechanical Engineering & Automation

基　　金：广东工业大学教育教学改革项目教育信息化教改项目(广工大教字(2022)59号)。

摘　　要：针对现有工程图样检测方法在处理手绘图像时面临精度不足和计算复杂度高的问题,提出了一种基于ResNext-Vit的手绘工程图样检测方法。该方法融合了ResNext和Vision Transformer特征提取模型,并提出了基于边缘检测的图像预处理模块。通过孪生神经网络和对比式学习技术,在有限的训练样本条件下,实现高精度多类别手绘工程制图的区分。经过自建工程图样数据集(HFJH)的严格实验验证,该方法取得了84.7%的准确率,实现了手绘工程图样的高效检测。In view of the challenges of insufficient accuracy and high computational complexity of existing engineering drawing detection methods in processing hand-painted images,this paper proposes a hand-painted engineering drawing detection method based on resnext VIT.This method combines resnext and vision transformer feature extraction model,and proposes an image preprocessing module based on edge detection.Through the twin neural network and comparative learning technology,under the condition of limited training samples,the high-precision multi category hand-painted engineering drawing can be distinguished.Through the strict experimental verification of self built engineering drawing data set(HFJH),the accuracy of this method is 84.7%,and the efficient detection of hand drawn engineering drawings is realized.

关 键 词：手绘工程制图 ResNext Vision Transformer 高效检测方法 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]