黄期峰,曾志强,洪智勇.基于机器视觉的卷纸包装检测模型设计[J].中国造纸学报,2023,38(4):85-91 |
![本文二维码信息](/download_upload_file.aspx?file_name=/uploadfile/two_dimensional_code_img/zzxb/2023/4/202304012.png) 二维码(扫一下试试看!) |
基于机器视觉的卷纸包装检测模型设计 |
Design of Roll Paper Packaging Detection Model Based on Machine Vision |
投稿时间:2022-08-23 |
DOI:10.11981/j.issn.1000-6842.2023.04.85 |
中文关键词: 卷纸包装 机器视觉 YOLOv4 深度学习 |
Key Words:roll paper packaging machine vision YOLOv4 deep learning |
基金项目:五邑大学港澳联合研发基金(2019WGALH21);广东省基础与应用基础研究基金(2020A1515011468);广东省普通高校特色创新类项目(2019KTSCX189)。 |
|
摘要点击次数: 879 |
全文下载次数: 3 |
中文摘要: |
针对卷纸包装检测效率低、人工成本高的问题,本研究基于机器视觉设计了一个卷纸包装检测模型,并命名为F-YOLOv4。首先利用工业相机在卷纸包装过程中采集目标图像,并人工标注制作成数据集;随后基于YOLOv4构建卷纸包装检测模型,通过引入轻量级的混合通道注意力模块,以强化重要特征同时避免背景噪声的引入;并设计了残差上采样模块以提升上采样的效果;最后在检测头部分,将不同分辨率的特征进行了融合以丰富特征图信息。研究结果表明,F-YOLOv4模型的准确率为97.53%,高于原始模型1.97%,检测速度为129 f/s,模型大小为39.7 MB。F-YOLOv4模型能够有效解决卷纸包装问题,为企业降低用人成本,提高生产效率。 |
Abstract: |
Aiming at the problems of low efficiency and high labor cost in roll paper packaging detection, a roll paper packaging detection model was designed based on machine vision and named F-YOLOv4. First, an industrial camera was used to collect target images in the process of roll paper packaging, and manually annotate them into a data set. Then, based on YOLOv4, a roll paper packaging detection model was built, and a lightweight mixed-channel attention module was introduced to enhance important features while avoiding the introduction of background noise. And the residual upsampling module was designed to improve the effect of upsampling. Finally, in the detection head part, the features of different resolutions were fused to enrich the feature map information. The experimental results showed that the accuracy of F-YOLOv4 was 97.53%, which was 1.97% higher than the original model, the detection speed was 129 f/s, and the model size was 39.7 MB. F-YOLOv4 can effectively solve the problem of roll paper packaging, reduce labor costs for enterprises, and improve production efficiency. |
查看全文 查看/发表评论 下载PDF阅读器 HTML |