[1]曾碧卿,杨 睿,李一娴,等.基于卷积神经网络的零件圆检测方法[J].计算机技术与发展,2023,33(11):64-71.[doi:10. 3969 / j. issn. 1673-629X. 2023. 11. 010]
　ZENG Bi-qing,YANG Rui,LI Yi-xian,et al.Part Circle Detection Method Based on Convolutional Neural Network[J].,2023,33(11):64-71.[doi:10. 3969 / j. issn. 1673-629X. 2023. 11. 010]

点击复制

基于卷积神经网络的零件圆检测方法()

分享到：

《计算机技术与发展》[ISSN:1006-6977/CN:61-1281/TN]

卷:

33

期数:

2023年11期

页码:

64-71

栏目:

媒体计算

出版日期:

2023-11-10

文章信息/Info

Title:

Part Circle Detection Method Based on Convolutional Neural Network

文章编号:

1673-629X(2023)11-0064-08

作者:

曾碧卿¹ ; 杨 睿¹ ; 2 ; 李一娴² ; 张雅蓉¹

1. 华南师范大学 软件学院,广东 佛山 528225;

2. 季华实验室,广东 佛山 528200

Author(s):

ZENG Bi-qing1 ; YANG Rui1 ; 2 ; LI Yi-xian2 ; ZHANG Ya-rong1

1. School of Software,South China Normal University,Foshan 528225,China;

2. Jihua Laboratory,Foshan 528200,China

关键词:

圆检测; 卷积神经网络; 目标检测; 语义分割; 霍夫变换

Keywords:

circle detection; convolutional neural network; object detection; semantic segmentation; Hough transform

分类号:

TP399

DOI:

10. 3969 / j. issn. 1673-629X. 2023. 11. 010

摘要:

在零件生产的场景中往往需要对零件产品进行质量检测,多孔零件中的圆孔是否符合生产标准也是质量检测中的重要环节。 为解决传统圆检测算法无法处理复杂场景下特定圆的检

测问题,该文设计了一种基于卷积神经网络的零件圆检测方法,将圆检测分为 3 个阶段,第 1 阶段使用 YOLOv5 目标检测模型对零件图片中的目标圆进行粗检测,将多圆检测问题简化

为单圆检测,获得含有单个目标圆的裁剪图片;第 2 阶段使用 BiSeNet 语义分割模型对单圆图片进行细检测,获得圆轮廓掩膜图;第 3 阶段使用改进的随机霍夫变换对圆参数进行检测,最终得到图中所有目标圆的半径与圆心坐标。经实验结果对比,该方法在多种阈值条件下的检测精度都高于其他对比方法,在 IoU 阈值为 0. 9 的情况下 F-meausre 达到96% ,能满足生产场景中的实时检测需求。

Abstract:

In the scene of parts production, it `s necessary to conduct quality inspection of parts products. Whether the round holes inporous parts meet the production standards is an important procedure in quality inspection. In order to solve the problem that traditionalcircle detection algorithms can hardly handle the detection of specific circles in complex scenes,we design a circle detection method basedon convolutional neural network. The circle detection is divided into three stages. The first stage uses the YOLOv5 object detectionmodel to perform rough detection on the target circle in the parts image,simplify the multi-circle detection problem into single-circle detection,and obtain a cropped?

image containing a single target circle. The second stage uses the BiSeNet semantic segmentation model toperform fine detection on the single-circle image,and obtains the circle contour binary map. The third stage uses improved randomizedHough transform to detect the circle parameters,and finally obtains the radius and center coordinates of target circles in the images. Aftercomparing the experimental results,the detection accuracy of the proposed method is higher than that of other comparison methods undervarious threshold conditions,when IoU threshold is 0. 9,F-meausre reaches 96% ,which can meet the real-time detection requirements inproduction scenes.

常用功能