基于深度卷积神经网络的异型纤维识别  

Profiled Fiber Image Recognition Based onDeep Convolutional Neural Network

作　　者：孙硕磊 吴欢 铁铮 Shuolei Sun;Huan Wu;Zheng Tie(College of Computer Science and Technology,Donghua University,Shanghai 201620,China;Songjiang Entry-Exit Inspection and Quarantine Bureau,Shanghai 201620,China#Email： [e-mail])

机构地区：[1]东华大学计算机科学与技术学院,上海201620 [2]松江出入境检验检疫局,上海201620

出　　处：《电气工程与自动化（中英文版）》2016年第2期50-55,共6页Electrical Engineering and Automation

基　　金：受基于约束条件的非负矩阵分解算法及其在纤维自动识别中的应用研究国家自然科学基金支持资助（61472075）

摘　　要：近年来,深度卷积神经网络作为深度学习模型之一被广泛应用于图像识别领域,不仅提高了图像识别的准确率,而且可以逐层自动地进行特征学习和提取,避免了传统识别算法中复杂的特征提取过程.针对异型纤维的分类问题,本文研究并设计了基于深度卷积神经网络的异型纤维识别方法.通过对5种异型纤维的分类实验表明,该方法平均识别率可达94.4%,较SVM分类器在识别精度上取得了显著的提高.In recent years, deep convolutional neural network （DCNN）, a deep learning method, has been widely used in the field of imagerecognition. It not only significantly improves the recognition accuracy, but also can automatically learn and extract features layerby layer to avoid complex feature extraction process of the traditional recognition algorithm. This paper firstly introduces theDCNN for the profiled image recognition, and designs a fiber image recognition method based on a proposed DCNN. Theclassification results of five types of profiled fibers show that the average recognition rate is 94.4% and significant improvementsare achieved in recognition accuracy as compared with SVM classifier.

关 键 词：深度学习 深度卷积神经网络 支持向量机 异型纤维 纤维识别 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]