摘要
将深度学习的图像识别应用到工业生产中是一个重要的应用方向.相比传统图像处理,深度学习在图像识别中具有高识别率、抗干扰性强等特点.首先采用小波变换对图像去噪、归一化,然后利用多层卷积对图像进行特征提取并采用全连接层和softmax分类器进行分类实现图像识别.在铝厂工业自动浇注过程中,对已经浇注完成和未完成的图像进行识别、解决传统图像处理在工业生产中多干扰、亮度不足的情况下难以识别的问题.实验结果表明,采用小波变换与深度学习融合对图像进行识别的识别率可达到91.88%,基本能满足铝厂工业生产的需要.
It is an important applicational direction to apply deep learning to image recognition in industrial production.Compared with traditional image processing,deep learning has the characteristics of high identification rate and strong anti-interference ability in image recognition.In the beginning,wavelet transformation is used to denoise and normalize the image.Then,multi-layer convolution is used to extract the image’s features.In the end,the whole connective layer and softmax classifier are used to finish recognizing image.In the automatic pouring process of aluminum plant industry,the finished and unfinished images are identified.This method solves the problem that traditional image processing is difficult to identify in the case of multiple interference and insufficient brightness in industrial production.The experimental results show that the recognition rate of image recognition using wavelet transform and deep learning fusion can reach 91.88%.
作者
易佳明
胡小龙
YI Jiaming;HU Xiaolong(School of Computer Science, Central South University, Changsha 410075, China)
出处
《湖北大学学报（自然科学版）》
CAS
2020年第3期320-324,共5页
Journal of Hubei University：Natural Science
关键词
工业生产
图像识别
深度学习
小波变换
industrial production
image recognition
deep learning
wavelet transform
