货车螺栓丢失故障图像识别算法研究

发布时间：2018-05-16 01:21

本文选题：故障图像识别 + 自适应中值滤波　；参考：《哈尔滨工业大学》2016年博士论文

【摘要】：随着我国铁路事业的高速发展,传统的人工巡检模式已经不能满足日益增长的货车安检工作需求。近年来,货车运行故障动态图像检测(TFDS)系统的投入使用,极大地提高了车检作业质量和效率。该系统可以识别货车关键部位图像的故障状态,有着重要的工程应用价值和理论研究意义。目前国内外在货车故障图像识别领域开展的研究工作相对较少,应用的技术也较为传统。针对货车螺栓部位的故障图像识别问题,本文建立了一套有效可靠的货车螺栓丢失故障图像识别系统,对其算法实现过程中的预处理、特征提取和分类识别等环节的相关技术进行了研究。本文主要包括以下四个方面的内容:针对预处理环节的滤波降噪问题,提出了改进自适应中值滤波(IRAMF)算法。设计了多级噪声检测策略用于准确地鉴别噪声,包括全局检测、局部检测、邻域相似度检测和边角检测等方法;设计了自适应滤波策略用于有效地滤除噪声,包括局部纹理走向子窗口合成滤波、曼哈顿距离加权均值滤波和最小有效窗口均值滤波等方法。IRAMF算法可以根据噪声点局部邻域内的噪声密度状态和纹理分布情况,自适应地调整滤波窗口的尺寸、形状和输出结果的计算方法。选取了4幅现场图像进行实验,结果表明该算法具有优秀的滤波性能,能够更好地保护纹理信息。针对特征提取环节,提出了完整方向局部二值模式(CDLBP)算子。设计了局部纹理信息编码策略用于提取图像局部差分结果的符号和幅值变化信息;设计了全局比较信息编码策略用于提取局部区域与图像全局的灰度和幅值均值的比较信息。这样可以提高特征向量的信息含量,增强区分能力。同时提出了分类加权排序模式选择(LWR-DLBP)方法,综合考虑了不同模式的出现概率、分布的一致性和类别信息的影响,这就降低了特征向量中的信息冗余。选取货车心盘螺栓丢失故障图像作为实验对象,测试了CDLBP算子和LWR-DLBP方法的性能,实验结果表明它们提取的特征向量具有优秀的区分能力。针对分类识别过程中的多参数优化问题,提出了改进师生交流优化(ITLBO)算法。将种群更新的贪婪策略改进为优选策略;设计了自适应教师教学阶段,包括自适应学习步长、自适应知识差距和教师解的Lévy学习等策略;设计了自适应学生交流阶段,包括自主学习、交流学习和补课学习等策略。这样可以加快算法的收敛速度,提高其全局探索能力。以固定维和可变维的标准测试函数为对象,将ITLBO算法与其他11种算法进行对比实验,结果表明该算法拥有更快的收敛速度和更高的寻优精度。对货车螺栓丢失故障图像识别算法的设计与应用进行了研究。介绍了TFDS系统的组织结构和技术要求。给出了货车故障图像识别算法的设计方案:首先应用CDLBP算子提取原始图像在不同尺度和方向下的Gabor响应图像的纹理特征,然后使用LWR-DLBP方法对各个通道的原始特征进行优化,接着对每个通道分别进行SVM分类识别,最后利用ITLBO算法对不同通道的预测标签分配适宜的权重,进而获得了最终的故障判别结果。测试了系统在识别心盘螺栓丢失故障、钩尾扁销螺栓丢失故障和安全链脱落故障时的性能表现,实验结果表明本文算法可以有效地识别货车螺栓部位图像的故障状态。
[Abstract]:With the rapid development of China's railway industry, the traditional manual inspection mode has not been able to meet the increasing demand for the safety inspection of freight cars. In recent years, the dynamic image detection (TFDS) system has greatly improved the quality and efficiency of the vehicle inspection operation. The system can identify the malfunction of the key parts of the freight car. State, it has important engineering application value and theoretical research significance. At present, there are relatively few research work in the field of vehicle fault image recognition at home and abroad, and the applied technology is more traditional. In this paper, a set of effective and reliable identification of truck bolt loss fault image recognition is established for the problem of fault image recognition of freight car bolt parts. In this paper, the related technologies of pre processing, feature extraction and classification recognition are studied. This paper mainly includes the following four aspects: the improved adaptive median filter (IRAMF) algorithm is proposed for the filtering and noise reduction problem of the preprocessing link. A multi level noise detection strategy is designed for accuracy. To identify the noise, including global detection, local detection, neighborhood similarity detection and edge angle detection, the adaptive filtering strategy is designed to filter the noise effectively, including local texture to sub window synthetic filtering, and the.IRAMF algorithm can be based on the method of the.IRAMF distance weighted mean filtering and the minimum effective window mean filtering. The noise density state and texture distribution in the local neighborhood of noise are adapted to adjust the size, shape and output of the filter window adaptively. 4 field images are selected to carry out experiments. The results show that the algorithm has excellent filtering performance and can better protect the texture information. The local two value mode (CDLBP) operator is used in the whole direction. The local texture information coding strategy is designed to extract the symbol and amplitude change information of the local difference results of the image, and the global comparison information coding strategy is designed to extract the comparison information of the gray and amplitude mean values of the local region and the image global. This can improve the feature vector. At the same time, the classification weighted sorting model selection (LWR-DLBP) method is proposed, which comprehensively considers the occurrence probability of different modes, the consistency of the distribution and the influence of category information. This reduces the information redundancy in the feature vector. It selects the lost fault image of the truck's heart disk bolt loss as the experimental object and tests the test. The performance of the CDLBP operator and the LWR-DLBP method, the experimental results show that the extracted feature vectors have excellent distinguishing ability. In view of the multi parameter optimization problem in the classification and recognition process, an improved teacher student communication optimization (ITLBO) algorithm is proposed. The greedy strategy of the population updating is improved to the optimal strategy, and the adaptive teacher teaching stage is designed. It includes the strategies of adaptive learning step, adaptive knowledge gap and teacher's L e vy learning, and designs adaptive student communication stages, including autonomous learning, exchange learning and lesson learning, so as to speed up the convergence speed of the algorithm and improve its global exploration ability. The ITLBO algorithm is compared with the other 11 algorithms. The results show that the algorithm has faster convergence speed and higher optimization accuracy. The design and application of the algorithm for identifying the fault image of the truck bolt loss are studied. The organization structure and technical requirements of the TFDS system are introduced. The recognition algorithm of the truck fault image is given. The CDLBP operator is used to extract the texture features of the Gabor response image of the original image in different scales and directions. Then the LWR-DLBP method is used to optimize the original features of each channel. Then, each channel is classified by SVM classification. Finally, the ITLBO algorithm is used to allocate the predictive labels for different channels. The results of the final fault discrimination are obtained, and the performance of the system is tested to identify the failure of the bolt loss, the loss of the bolt and the tail bolt and the failure of the safety chain. The experimental results show that the algorithm can effectively identify the fault status of the bolt position image of the freight car.

【学位授予单位】：哈尔滨工业大学
【学位级别】：博士
【学位授予年份】：2016
【分类号】：U279.3;TP391.41

【相似文献】