摘要
高频元件精密加工工件需要定制,难以获取足量的图像样本以训练传统的CNN模型来完成自动化分类。提出基于元学习训练策略的关联非局部网络高频工件分类模型。在经典CNN模型的基础上引入NLN模块,通过关联样本的全局和局部特征提取自适应的任务特征;通过最大化互信息约束优化模型,提高模型的鲁棒性。实验结果表明:与多种主流的小样本分类模型相比,所提模型对miniImageNet和高频工件数据集分类的准确率有显著提高。
High-frequency precise machining components need to be customized,which leads to the difficulty in obtaining enough image samples to train traditional CNN models for the realization of automatic classification.This paper proposes a related non-local network high-frequency components classification model based on a meta-learning training strategy.A NLN module is introduced based on the classical CNN model to extract adaptive task features by correlating global and local features of samples.The model is optimized by maximizing mutual information constraints to improve the robustness of the model.The experimental results show that the proposed model significantly improves the classification accuracy on miniImageNet and high-frequency precise components datasets compared with various mainstream few-shot learning classification models.
作者
许知建
欧阳
李毅
李柏林
熊鹰
XU Zhijian;OU Yang;LI Yi;LI Bailin;XIONG Ying(School of Mechanical Engineering,Southwest Jiaotong University,Chengdu 610031,China)
出处
《机械制造与自动化》
2023年第6期89-93,101,共6页
Machine Building & Automation
基金
四川省重大科技专项(18ZDZX0140)。
关键词
高频工件
元学习
非局部网络
互信息约束
图像分类
high-frequency component
meta-learning
non-local network
mutual information constraints
image classification