基于多核SVM的绝缘子视觉识别技术研究

发布时间：2018-08-18 17:03

【摘要】：研究绝缘子视觉识别系统,实现铁路接触网的绝缘子的智能化清洗,对排除污闪隐患、保障铁路交通系统稳定运行具有重要意义。本文通过采用基于核函数的支持向量机算法(Support Vector Machine,SVM)建立绝缘子视觉识别模型,实现了绝缘子图像样本的识别检测。主要研究内容如下:(1)采集绝缘子图像样本并进行二值化及ROI区域提取,通过Hu、LBP、H O G、H a a r等纹理特征描述算子建立绝缘子特征样本,采用归一化和主成分分析方法建立训练测试样本集。(2)建立基于线性核、径向基核、多项式核的SVM训练模型,以及基于线性核SVM方法的单纹理类型训练模型,通过对比分析各模型的用时及识别性能,验证了基于核函数的SVM方法对绝缘子识别的有效性,确定了多核绝缘子视觉识别模型组合形式。(3)分析基于核函数的SVM方法中参数对最终识别模型的性能影响,将模型识别准确率和支持向量个数作为模型性能的综合评价指标,并提出了相应的评价函数;在分析多种群体智能优化算法基础上,采用差分进化算法对SVM核函数方法的参数进行寻优,进一步提高了绝缘子识别模型的性能。(4)采用六角晶格标定板和SVM预测算法建立具有全视域采样的绝缘子双目视觉空间定位系统,通过实验分析映射模型的空间定位精度验证了模型的有效性。
[Abstract]:It is of great significance to study the visual identification system of insulators and realize the intelligent cleaning of insulators in railway catenary. It is of great significance to eliminate the hidden trouble of pollution flashover and to ensure the stable operation of railway traffic system. In this paper, the support vector machine (Support Vector machine) algorithm based on kernel function is used to establish the insulator visual recognition model, which realizes the detection of insulator image samples. The main research contents are as follows: (1) the insulator feature samples are collected and binary and ROI region are extracted, and the insulator feature samples are established by using the texture feature description operator such as Hugh LBP, H O O Go H a a r, etc. The SVM training model based on linear kernel, radial basis kernel, polynomial kernel and single texture type training model based on linear kernel SVM method are established. The effectiveness of the kernel function based SVM method for insulator identification is verified by comparing the time use and recognition performance of each model. The combination form of multi-core insulator visual recognition model is determined. (3) the influence of parameters on the performance of the final identification model in the kernel function based SVM method is analyzed. The accuracy of model recognition and the number of support vectors are taken as the comprehensive evaluation indexes of the model performance. Based on the analysis of various swarm intelligence optimization algorithms, the differential evolution algorithm is used to optimize the parameters of the SVM kernel function method. The performance of insulator identification model is further improved. (4) using hexagonal lattice calibration board and SVM prediction algorithm, a binocular visual location system for insulator with full horizon sampling is established. The validity of the model is verified by the experimental analysis of the spatial positioning accuracy of the mapping model.
【学位授予单位】：西南科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【相似文献】