农业装备与车辆工程2025，Vol.63Issue(6)：124-128,5.DOI:10.3969/j.issn.1673-3142.2025.06.022
基于迁移学习的机械零件识别研究
Research on mechanical part recognition based on transfer learning
摘要
Abstract
To achieve accurate and efficient mechanical part recognition,a transfer learning-based part identification method was proposed.By reusing the pre-trained parameters of ResNet50 on ImageNet,combined with the feature extraction capabilities and gradient vanishing resistance advantages of the residual network,the underlying convolutional layer parameters were retained while the fully connected layer was replaced.The model was fine-tuned on a self-built dataset containing 44 208 images of 55 types of parts.Experimental results showed that the model achieved an average recognition accuracy of 99.85%on the test set,validating the effectiveness of transfer learning in small-sample scenarios.Meanwhile,an intelligent recognition system interactive interface for the web page has been developed,enabling human-computer interaction functions such as taking photos and recognizing parts.This research provides a reference for image-based automated part identification.关键词
机械零件识别/图像/迁移学习/ResNet50/交互界面Key words
mechanical part recognition/image/transfer learning/ResNet50/interactive interface分类
信息技术与安全科学引用本文复制引用
尹晨旭,周扬..基于迁移学习的机械零件识别研究[J].农业装备与车辆工程,2025,63(6):124-128,5.基金项目
西安航空学院2024年省级大学生创新创业训练项目"基于深度学习的智能零件识别系统"(S202411736112) （S202411736112）
陕西省自然科学基础研究计划项目(2025JC-YBQN-545) （2025JC-YBQN-545）
