基于改进UNet网络的金丝球焊直径测量研究  

Research on diameter measurement of gold wire ball solder joints based on improved UNet network

作　　者：张杰[1] 唐立新[1] 陈子章 安成 ZHANG Jie;TANG Lixin;CHEN Zizhang;AN Cheng(School of Mechanical Science and Engineering,Huazhong University of Science and Technology,Wuhan 430074,China)

机构地区：[1]华中科技大学机械科学与工程学院,武汉430074

出　　处：《现代制造工程》2022年第5期110-114,共5页Modern Manufacturing Engineering

摘　　要：提出了一种基于改进UNet网络的金丝球焊焊点精确分割与测量的新方法。改进UNet网络由编码器和解码器两部分构成,其中编码器主要用来提取图像特征,使用在ImageNet数据集上预训练分类网络的卷积模块权重初始化编码器部分,可以在不增加训练数据的情况下,加速网络训练且避免过拟合;解码器主要是结合深层特征和浅层特征以实现精确分割,使用改进的多尺度卷积模块替换原网络中的单尺度卷积模块,使解码器能综合利用不同感受野的特征,进一步提升网络的分割精度。实验结果表明,改进UNet网络与原始UNet网络相比,其测试集分割交并比和F分数分别提升了2.04%和1.58%,且直径测量平均误差从7.734μm降低到1.435μm,满足实际检测需求。A new method for precise segmentation and measurement of gold wire ball solder joints based on improved UNet network was proposed.The improved UNet network was composed of an encoder and a decoder.The encoder of the network was used to extract the shallow-level and deep-level image features.In order to improve the training speed and avoid overfitting, the weights of the convolution module of the classification network, which had been trained on the ImageNet data set, were used to initialize the network encoder.The network decoder performs accurate semantic segmentation of solder joints in combining the shallow-level and deep-level features of the images.An improved multi-scale convolution module was used to replace the single-scale one in the original network decoder, which enables the decoder to comprehensively utilize the image features from different receptive fields and further improve the segmentation accuracy of the network.The experimental results show that, compared with the original network, intersection over union and F-score of the test set of the improved network are improved by 2 % and 1.5 % respectively, and the average diameter measurement error is reduced from 7.734 μm to 1.435 μm, which meets the actual detection requirements.

关 键 词：UNet 金丝球焊 直径测量 半导体器件 视觉检测 

分 类 号：TN389[电子电信—物理电子学] TP242.2[自动化与计算机技术—检测技术与自动化装置]