查询结果:   周忠义,吴谨,朱磊.基于多路特征融合和深度学习的露霜图像分类[J].计算机应用与软件,2018,35(10):205 - 210,231.
中文标题
基于多路特征融合和深度学习的露霜图像分类
发表栏目
图像处理与应用
摘要点击数
518
英文标题
DEW AND FROST IMAGE CLASSIFICATION BASED ON MULTI-PATH FEATURE FUSION AND DEEP LEARNING
作 者
周忠义 吴谨 朱磊 Zhou Zhongyi Wu Jin Zhu Lei
作者单位
武汉科技大学信息科学与工程学院 湖北 武汉 430081     
英文单位
College of Information Science and Engineering, Wuhan University of Science and Technology, Wuhan 430081, Hubei, China     
关键词
露霜图像分类 多路特征融合 深度学习 ResNet-50模型
Keywords
Dew and frost image classification Multi-path feature fusion Deep learning ResNet-50 model
基金项目
国家自然科学基金青年项目(61502358)
作者资料
周忠义,硕士生,主研领域:模式识别。吴谨,教授。朱磊,副教授。 。
文章摘要
针对基于特征变化进行露霜图像识别中特征提取难度较大的问题,提出一种基于多路特征融合和深度学习的露霜图像分类方法。利用传统的卷积神经网络对露霜图像进行分类;在ResNet-50模型基础上,通过增加浅层网络层到深层网络层的多个通路,将具有更强细节纹理信息的浅层特征和具有更明确语义分类信息的深层特征相结合,增强后续卷积运算的特征信息。在露霜自动化观测设备采集的图像集上测试,实验结果表明,该方法的分类准确率达到99.2%。
Abstract
Aiming at the difficulty of feature extraction in dew and frost image recognition based on feature change, we proposed a dew-frost image classification method based on multi-path feature fusion and deep learning. The traditional convolutional neural network was used to classify the dew and frost images. On the basis of ResNet-50 model, we combined the shallow features with more detailed texture information and the deep features with more specific semantic classification information to enhance the feature information of subsequent convolution operations by adding multiple paths from the shallow network layer to the deep network layer. We tested it on the image set collected by dew and frost automatic observation equipment. The experimental results show that the classification accuracy of the proposed method reaches 99.2%.
下载PDF全文