查询结果:   石陆魁,党磊,杨璐,师胜利.融合流形特征的路面破损图像识别方法[J].计算机应用与软件,2016,33(2):150 - 152,196.
中文标题
融合流形特征的路面破损图像识别方法
发表栏目
人工智能与识别
摘要点击数
860
英文标题
RECOGNITION METHOD OF PAVEMENT DISTRESS IMAGES FUSING MANIFOLD FEATURES
作 者
石陆魁 党磊 杨璐 师胜利 Shi Lukui Dang Lei Yang Lu Shi Shengli
作者单位
河北工业大学计算机科学与软件学院 天津 300401 河北师范大学信息技术学院 河北 石家庄 050024    
英文单位
School of Computer Science and Engineering,Hebei University of Technology,Tianjin 300401,China School of Information Technology,Hebei Normal University,Shijiazhuang 050024,Hebei,China    
关键词
流形学习 路面破损图像识别 特征融合 拉普拉斯特征映射法
Keywords
Manifold learning Pavement distress image recognition Feature fusion Laplacian eigenmaps
基金项目
作者资料
石陆魁,副教授,主研领域:机器学习,数据挖掘。党磊,硕士。杨璐,讲师。师胜利,讲师。 。
文章摘要
研究路面破损图像识别的特征提取优化问题。为了克服常见的破损密度因子或坐标轴投影等特征提取时易受噪声影响,仅从底层视觉角度描述破损图像裂缝特性,不能高效、精确地区别不同裂缝的问题,提出一种融合流形特征的路面破损识别方法。首先利用流形学习中的Laplacian Eigenmaps算法提取图像的低维流形特征,令其作为图像裂缝的高层语义,然后将流形特征与破损密度因子或坐标轴投影等底层视觉特征融合,利用融合后的特征识别裂缝类别。仿真结果表明,将流形特征与其他特征融合后,可以从高层语义、底层视觉两个层面全方位的描述路面裂缝,极大地提高路面裂缝的识别精度。
Abstract
This paper mainly discusses the feature extraction optimisation for pavement distress images recognition. In order to overcome the common problems that the features such as damage density factors or coordinate axes projections are sensitive to noise, and it cannot effectively and accurately distinguish various pavement cracks by describing the cracks feature of distress images from underlying visual perspective only, we presented a pavement distress recognition method fusing the manifold features. In the method, first it uses Laplacian eigenmaps algorithm in manifold learning to extract the low dimensional manifold features from pavement images and makes it as the high-level semantics of image cracks. Then it fuses the manifold features with the underlying visual features of distress density factors or coordinates projections, and utilise the fused features to recognise the category of cracks. Simulation results showed that with the fusion of manifold features and other features, it was able to give a full range description on pavement cracks from two dimensions of high-level semantics and underlying visual, and greatly improved the accuracy of pavement cracks recognition.
下载PDF全文