应用卷积网络及深度学习理论的羊绒与羊毛鉴别

发布时间：2018-03-30 10:53

  本文选题：羊绒　切入点：羊毛　出处：《纺织学报》2017年12期

【摘要】：为解决羊绒与羊毛纤维图像难以鉴别的问题,提出一种基于卷积网络和深度学习理论的鉴别方法。使用sigmoid分类器将卷积深度网络提取的纤维图像特征进行粗分类,根据验证集合验证结果并记录网络的最优权重。根据整体的分类网络所获取的权值,对每张样本图片使用改进的局部增强整体的网络模型提取局部特征,并对局部特征和整体特征进行融合,根据这些融合特征建立新的分类网络。在此基础上,使用鄂尔多斯标准羊绒与羊毛数据集对网络进行50轮次的迭代训练,得到的最优准确率达92.1%。实验结果表明:采用深度卷积网络表征纤维,并对羊绒羊与毛纤维图像进行分类的方法,能够有效解决羊绒、羊毛等类似纤维鉴别问题;若用于商业检测,还需更多数据集的验证。
[Abstract]:In order to solve the problem that the image of cashmere and wool fiber is difficult to distinguish, a method based on convolution network and depth learning theory is proposed. The feature of fiber image extracted by convolution depth network is roughly classified by sigmoid classifier. According to the verification result of the verification set and the optimal weight of the network, according to the weights obtained by the whole classification network, the improved local enhancement global network model is used to extract the local features for each sample picture. Based on the fusion of local and global features, a new classification network is established. On this basis, 50 rounds of iterative training of the network are carried out by using the Ordos standard cashmere and wool data sets. The experimental results show that the method of using deep convolution network to characterize the fiber and classify the image of cashmere and wool fiber can effectively solve the problem of identifying similar fibers such as cashmere wool and so on. If used for commercial detection, more data sets need to be validated.
【作者单位】： 东华大学纺织学院;东华大学纺织面料技术教育部重点实验室;
【分类号】：TP391.41;TS102.31
，

本文编号：1685608