基于HOG特征与SVM的胶体气泡识别方法研究

Research on the identification method of colloid bubble based on HOG feature and SVM

本文提出了一种基于HOG特征与支持向量机的分类方法,来代替目前人工检测光纤连接器中的胶体是否存在气泡。该方法先对视频中的胶体部分进行逐帧取样,根据样本制作正、负样本训练集以及测试集,接着选择适当的参数提取了训练集和测试集中胶体气泡HOG特征,并通过支持向量机来进行训练和测试统计的HOG特征,来识别胶体是否存在气泡。最后以人工识别为标准,并与本方法的分类结果进行验证。其实验结果表明该方法能够有效地将有、无气泡的样本进行分类,并在时间和人力上取得明显的优势。

In this paper,a classification method based on HOG feature and Support Vector Machine is proposed to replace manual method to detect colloid bubble in the optical fiber connector. Firstly,this method gets the image about colloidal part from video frame by frame,making positive and negative training sample set and test set according to the sample. Then,the appropriate parameters are selected to extract colloid bubble HOG features from training set and test set,and HOG features trained and tested by the Support Vector Machine（ SVM） to detect whether there is a bubble. Finally,the classification results of this method are verified by artificial recognition. The results showthat this method can effectively classify the samples without bubbles and gain obvious advantages in time and manpower.