| 李欢,李亮.基于卷积神经网络与机器视觉的纸张尘埃度测量系统的设计与应用研究[J].中国造纸,2025,(8):157-163 |
二维码（扫一下试试看！） |
| 基于卷积神经网络与机器视觉的纸张尘埃度测量系统的设计与应用研究 |
| Design and Research of Paper Dirt Determination System Based on Convolutional Neural Network and Machine Vision |
| 收稿日期：2025-01-23 修订日期：2025-04-06 |
| DOI：10.11980/j.issn.0254-508X.2025.08.020 |
| 关键词: 纸张尘埃度 卷积神经网络（CNN） 机器视觉 图像处理 |
| Key Words:paper dirt convolutional neural network (CNN) machine vision image processing |
| 基金项目: |
|
| 摘要点击次数: 2159 |
| 全文下载次数: 1248 |
| 摘要:本研究基于卷积神经网络（CNN）与机器视觉，设计了纸张尘埃度测量系统。该系统基于模型训练和检验2个模块构建，使用高分辨率扫描仪获取尘埃数据集和纸张样品图片，使用不同优化算法训练分类模型，并采用对角线测量算法，制作标准尘埃像素表用于定级和分类统计，进而计算尘埃度。结果表明，该系统的精度可达0.007 mm2，优于GB/T 1541—2013《纸和纸板 尘埃度的测定》要求，分类准确度达95.89%，能够实现多类纸品的全量程测量，单样本重复性测量误差为0，相比人工检测系统单样本检测用时缩短了约97%，实现了纸类产品尘埃度的高效、精准检测。 |
| Abstract:This study designed a paper dirt determination system based on convolutional neural network (CNN) and machine vision. The system was constructed with two modules， model training and testing. High-resolution scanners were used to obtain dirt datasets and images of paper samples. Different optimization algorithms were applied to train the classification model， and a diagonal measurement algorithm was adopted. A standard dirt pixel table was created for grading and classification statistics， thereby calculating the dirt. The results showed that the precision of the system could reach 0.007 mm²， which was better than the requirement specified in GB/T 1541—2013 “Paper and board—Determination of dirt”. The classification precision reached 95.89%， enabling full-range determination of various paper products. The repeatability determination error of a single sample was 0. Compared with manual detection， the single-sample detection testing time of the system was reduced by about 97%， realizing efficient and accurate detection of dirt in paper products. |
| 查看全文
HTML
查看/发表评论 下载PDF阅读器 |