Computer Engineering and Applications ›› 2022, Vol. 58 ›› Issue (7): 213-219.DOI: 10.3778/j.issn.1002-8331.2104-0299

• Graphics and Image Processing • Previous Articles     Next Articles

Defect Detection of Chip on Carrier Based on Lightweight Convolutional Neural Network

ZHOU Tianyu, ZHU Qibing, HUANG Min, XU Xiaoxiang   

  1. 1.Key Laboratory of Advanced Process Control for Light Industry, Ministry of Education, Jiangnan University, Wuxi, Jiangsu 214122, China
    2.Wuxi CK Electric Control Equipment Co., Ltd., Wuxi, Jiangsu 214400, China
  • Online:2022-04-01 Published:2022-04-01

基于轻量级卷积神经网络的载波芯片缺陷检测

周天宇,朱启兵,黄敏,徐晓祥   

  1. 1.江南大学 轻工过程先进控制教育部重点实验室,江苏 无锡 214122 
    2.无锡市创凯电气控制设备有限公司,江苏 无锡 214400

Abstract: Chip on carrier(COC) is an important component of transmitter optical subassembly(TOSA) and is widely used in the field of optical communication to realize photoelectric conversion. Aiming at the problem of real-time detection of three different types of defects on COC, such as collapse, positioning column damage and waveguide stain, a defect detection algorithm of COC based on lightweight convolution neural network, YOLO-Efficientnet, is proposed. Firstly, in order to reduce network parameters and shorten detection time, lightweight convolution neural network, efficientnet, is used for image feature extraction as a backbone network. On the basis of MBConv, the attention idea of SENet is introduced, and the attention mechanism is introduced in the channel dimension. Secondly, in order to solve the problem of information loss during the downsampling process, the spatial pyramid pooling(SPP) structure is introduced to increase the receptive field of the image and separate more significant context features. Finally, to solve the problems of multi-scale COC defects and difficult detection of small targets in waveguide region, PANet structure is introduced for multi-scale feature fusion. Results of the experiments show that the algorithm proposed in this paper has an accuracy of 98.5% for COC defect detection, and the detection time reaches 0.42 seconds per picture, meeting the requirements of real-time detection.

Key words: computer vision, defect detection of chip on carrier, object detection, YOLO-Efficientnet, transmitter optical subassembly

摘要: 载波芯片(chip on carrier,COC)是光发射次模块(transmitter optical subassembly,TOSA)的重要组成部分,被广泛应用于光通信领域,实现光电转换。针对载波芯片崩口、定位柱破损以及波导污渍三种不同类别缺陷的实时检测问题,提出了一种基于轻量级卷积神经网络的载波芯片缺陷检测算法YOLO-Efficientnet。为了减少网络参数,缩短检测时间,采用轻量级卷积神经网络Efficientnet作为主干网络对图像进行特征提取,在移动翻转瓶颈卷积(MBConv)的基础上,引入了压缩与激发网络(SENet)的注意力思想,在通道维度上引入注意力机制;为了解决下采样的过程中导致信息丢失的问题,引入空间金字塔池化(SPP)结构来增大图像的感受野,分离出更加显著的上下文特征。针对COC缺陷多尺度以及波导区域污渍小目标难以检测的问题,引入了PANet结构进行多尺度特征融合。实验结果表明,提出的算法对COC缺陷检测的准确率达到了98.5%,检测时间达到每张图片0.42?s,满足实时检测的需求。

关键词: 计算机视觉, 载波芯片缺陷检测, 目标检测, YOLO-Efficientnet, 光发射次模块