用于卷积神经网络图像预处理的目标中心化算法

来源期刊:中南大学学报(自然科学版)2019年第3期

论文作者:吴爱国 董秋成 董娜 冯伟

文章页码:579 - 587

关键词:零件;识别;卷积神经网络;数据增强;中心化;目标提取

Key words:parts; recognition; convolutional neural network; data augmentation; centralization; object extraction

摘    要:为解决工业生产中对不同零件进行自动分类的问题,提出一种基于卷积神经网络的模式识别算法,对29种不同尺寸的螺丝、螺母和垫片进行分类。首先采集待分类零件的图像数据,通过数据增强得到数据集,然后设计一种简化的卷积神经网络。提出一种对图像中的目标位置进行中心化的图像预处理算法,它能够提取图像中目标所在的区域并将其移动到图像中心位置。研究结果表明,与不采用目标中心化算法的传统方法相比,总体准确率从97.59%提升至99.96%,具有最低准确率的零件的准确率从85.83%提升至99.67%。使用卷积神经网络对背景纯净且目标明显的图像进行分类时,使用本文提出的目标中心化算法进行图像预处理能够显著提高网络的识别准确率。

Abstract: To solve the problem of classifying different parts automatically in industrial production, a pattern recognition algorithm based on convolutional neural network was raised and 29 different sizes of screws, nuts and washers were classified. Firstly, image data of the parts that were going to be classified were collected, and the dataset was created by data augmentation. Then, a simplified convolutional neural network was designed. An image preprocessing algorithm to centralize the position of the target in the image was raised, which can extract the target area in the image and move it to the center of the image. The results show that compared with traditional method without target-centralization algorithm, the total error is raised from 97.69% to 99.96, and the accuracy of the part which has the lowest accuracy is raised from 85.83% to 99.67%. When convolutional neural network is used to classify images which has pure background and obvious object, using the target-centralization algorithm raised in this paper to preprocess the images can improve the accuracy of the network significantly.

相关论文

  • 暂无!

相关知识点

  • 暂无!

有色金属在线官网  |   会议  |   在线投稿  |   购买纸书  |   科技图书馆

中南大学出版社 技术支持 版权声明   电话:0731-88830515 88830516   传真:0731-88710482   Email:administrator@cnnmol.com

互联网出版许可证:(署)网出证(京)字第342号   京ICP备17050991号-6      京公网安备11010802042557号