Scholarlink

<p>In recent years, visual inspection has played an increasingly important role in the assembly process of complex electronic equipment, particularly in scenarios involving multi-variety and small-batch production. However, visual inspection is limited by the availability of high-quality and diverse datasets. To obtain a large amount of synthetic data using few-shot real data, we propose a synthetic data generation method. Details are as follows: First, we use our designed network to extract material parameters of the parts by few-shot real data, and then adjust the material parameters based on the extracted lighting information. Then the restored parts are rendered diversely by a 3D graphics engine to generate a large amount of realistic synthetic data. Second, we proposed an improved object detection method by adding attention mechanisms and multi-scale feature fusion module based on R-FCN, effectively enhancing detection accuracy for parts of various sizes. Finally, we validated the method using a laboratory equipment model, which included training and testing with both real and synthetic data on various object detection methods. The experimental results show that the proposed method effectively performs quality inspection of the assembly process for large complex electronic equipment with multi-variety and small-batch, with a detection accuracy of 95.1%. The on-site verification by a company indicates that the research in this paper addresses the issue of assembly quality inspection for small-batch complex products, providing a practical solution for industrial automated assembly inspection.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A deep learning based visual inspection of small-batch electronic assembly using few-shot-driven synthetic data

  • 期刊论文

  • 发表日期：

    2025年4月26日

查看全文
  • Mingxing Jiang,
  • Tingyu Liu,
  • Songyang Li,
  • Xiao Lai,
  • Lei Jiao,
  • Zhonghua Ni

摘要

In recent years, visual inspection has played an increasingly important role in the assembly process of complex electronic equipment, particularly in scenarios involving multi-variety and small-batch production. However, visual inspection is limited by the availability of high-quality and diverse datasets. To obtain a large amount of synthetic data using few-shot real data, we propose a synthetic data generation method. Details are as follows: First, we use our designed network to extract material parameters of the parts by few-shot real data, and then adjust the material parameters based on the extracted lighting information. Then the restored parts are rendered diversely by a 3D graphics engine to generate a large amount of realistic synthetic data. Second, we proposed an improved object detection method by adding attention mechanisms and multi-scale feature fusion module based on R-FCN, effectively enhancing detection accuracy for parts of various sizes. Finally, we validated the method using a laboratory equipment model, which included training and testing with both real and synthetic data on various object detection methods. The experimental results show that the proposed method effectively performs quality inspection of the assembly process for large complex electronic equipment with multi-variety and small-batch, with a detection accuracy of 95.1%. The on-site verification by a company indicates that the research in this paper addresses the issue of assembly quality inspection for small-batch complex products, providing a practical solution for industrial automated assembly inspection.