基于旋转框的电子元器件检测  

Electronic Components Detection Based on Rotated Bounding Box

作　　者：汪威[1] 李琴锋 王冲 胡新宇[1] WANG Wei;LI Qin-feng;WANG Chong;HU Xin-yu(School of Mechanical Engineering,Hubei University of Technology,Wuhan 430068,China)

机构地区：[1]湖北工业大学机械工程学院,湖北武汉430068

出　　处：《仪表技术与传感器》2023年第3期33-38,49,共7页Instrument Technique and Sensor

基　　金：国家自然科学基金项目(61976083)。

摘　　要：针对印刷电路板上微小密集且方向任意分布的电子元器件检测过程中存在检测框与目标轮廓贴合度较差的问题,提出一种采用旋转框代替传统水平框的元器件检测方法。通过自建数据集并基于两阶段旋转目标检测算法对元器件进行检测,将特征提取网络改进为Swin Transformer后网络性能进一步提升,精度(mAP)达到98.23%,比原算法提高了1.52%。同时与5种旋转目标检测算法进行对比实验,文中方法检测效果均优于其他方法。A component detection method using a rotating bounding box instead of the traditional horizontal bounding box was proposed to address the problem of poor fit between the detection box and the object contour in the detection process of tiny and dense electronic components with arbitrary direction distribution on printed circuit boards.By constructing our own dataset and detecting components based on the two-stage rotated object detection algorithm,the performance of the network was further improved by replacing the feature extraction network with Swin Transformer,and the accuracy(mAP)reached 98.23%,which was 1.52%higher than the original algorithm.At the same time,comparison experiments with 5 kinds of rotated object detection algorithms are conducted,and the detection effect of this paper s method is better than other methods.

关 键 词：目标检测 深度学习 神经网络 印刷电路板 电子元器件 旋转框 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]