Low-level features in present railway fastener detection algorithms have poor descriptive ability and the error rates of these features are high.Aiming at these problems,this paper proposed a fastener detection model based on local features and semantic information.Firstly,it represented the profile information of fastener images by calculating low-level local features in the nonlinear spaces of images.Secondly,it eliminated the polysemy of visual words by dividing an image into four sub-ima-ges.The symmetry between the left and right part of a fastener and the similarity between the upper and lower part of a fastener caused the polysemy.Then,it constructed visual words according to the sub-images and acquired semantic information vectors via integrating the set of local features.Finally,it used these vectors to train classifier and detect fasteners.Rate of miss was 0.67% on balanced fastener samples.Experimental results show that the proposed algorithm has lower rates of miss and false alarm and stronger detect ability than existing detection methods.%针对现有底层特征识别扣件状态的算法存在描述能力差、错误率高等问题,提出一种基于扣件局部特征和语义信息的扣件检测模型。首先,在图像的非线性空间中计算扣件底层局部特征来表达扣件轮廓信息;然后,将图像分为四个子图,有效克服了由于扣件左右对称、上下相似造成的单词多义性问题;再根据扣件子图构造视觉单词,由底层特征整合得到语义信息向量;最后,以该向量训练分类器,判断待检扣件状态。对均衡的扣件样本进行测试,漏检率仅为0.67%。实验表明所提算法较现有方法,漏检率和误报率明显降低,检测能力增强。
展开▼